Home

Maui Administrator`s Guide - Adaptive Computing Documentation

Contents

1. NODESETATERIBUTE PROCSPEED NONE NOTE enabled in Maui 3 0 7 Maui will create node sets containing nodes with and higher See Node Set common processor speeds Overview specifies the length of time Maui will delay a job if adequate idle resources are NODESETDELAY 0 00 00 i wnat available but not adequate NODESETDELAY DD HH MM SS 0 00 00 resources within node set Maui will create node sets containing nodes with constraints NOTE enabled in common processor speeds Maui 3 0 7 and higher See Node Set Overview specifies the list of node NODESETPOLICY ONEOF attribute values which willbe NODESETATTRIBUTE FEATURE NODESETLIST STRING NONE considered for establishing node NODESETLIST switchA switchB lt gt pets NOTE enabled in Maur Maui will allocate nodes to jobs either using only 3 0 7 and higher See Node Set pots i nodes with the switchA feature or using only nodes Overview with the switchB feature specifies how nodes will be allocated to the job from the NODESETPOLICY ONEOF various node set generated NODESETATTRIBUTE NETWORK ae Maui will create node sets containing nodes with and higher See Node Set common network interfaces Overview specifies how resource sets will n when more as one INODESETPRIORITYTYPE BESTRESOURCE PASTO G TESOUTCE Call Call De NODESETATTRIBUTE PROCSPEED NODESETPRIORITYTYPE ge os MINLOSS found NOTE This par
2. JOBNAME USERNAME STATE PROC REMAINING STARTTIME fr28n13 709 0 dsheppar Running 1 0255209 Fri Aug 29 13227536 fr28n07 2303 0 dsheppar Running Al 0 55 10 Fri Aug 29 13 27 37 fr17n08 1349 0 dsheppar Running 1 1 02 29 Fri Aug 29 13 34 56 fr28n15 4355 0 dsheppar Running 1 1303208 Eri Aug 2g 13335335 fr28n05 2098 0 ebylaska Running 16 1 25 17 Fri Aug 29 11 57 45 f r28n05 2095 0 kossi Running I 1 26 24 Fri Aug 29 03 58 51 fr28n13 683 0 xztang Running 8 2 23 01 Thu Aug 28 17 52 08 fr28n15 4354 0 moorejt Running 16 3 41 06 Fri Aug 29 12 18 33 fr17n08 1341 0 mukho Running 8 3 41 48 Thu Aug 28 18 24 15 fr17n05 1393 0 zhong Running 8 4 01 47 Fri Aug 29 04 39 14 f r28n05 2097 0 zhong Running 8 4 50 03 Fri Aug 29 05 27 30 fr28n11 3080 0 mukho Running 8 5 12 21 Thu Aug 28 19 54 48 fr28n13 682 0 wengel Running 32 5 23 51 Thu Aug 28 19 56 58 fr28n05 2064 0 vertex Running 1 6 29 55 Thu Aug 28 23 02 22 fr28n11 3037 0 vertex Running 1 6 29 55 Thu Aug 28 23 02 22 fr28n09 26 0 rampi Running 1 8 37 27 Thu Aug 28 11 09 54 fr17n08 1328 0 vertex Running 1 9 29 49 Fri Aug 29 02 02 16 fr17n10 1467 0 kossi Running 1 10 27 10 Fri Aug 29 12 59 37 fr28n09 49 0 holdzkom Running 8 13 13 08 Fri Aug 29 11 45 35 fr17n07 1498 0 jpark Starting 16 14 10 05 Fri Aug 29 04 42 32 fr17n05 1384 0 zhong Running 8 18 45 27 Fri Aug 29 14 22 54 fr28n07 2300 0 Jimenez Running 16 18 54 12 Fri Aug 29 09 26 39 fr17n09 529 0 vertex Running al 19 03 49 Fri Aug
3. Class Zero of more of User System and Types of job holds currently Holds Batch applied to job Image Size Image Size KINTEGER gt gt Size Size of job data in MB Size of job data in MB data in MB as ee of real memory required Memory lt INTEGER gt per node in MB Network STRING gt Pa network adapter required Nodecount XINTEGER gt gt Number of nodes required by job of nodes Number of nodes required by job by job a o E lt STRING gt ah operating system required by Panition Mask ALL or colon delimited list of i of partitions the job has access partitions PE lt FLOAT gt Namber of processor equivalents requested by job Qos lt STRING gt oan of Service associated with QueueTime lt TIME gt LA job was submitted to resource management system Number of times job has been SanCout Sana started by Maui StartPriority StartPriority KINTEGER gt Start priority Start priority ofjob Start priority ofjob State EAEE of Idle Starting Running etc Current Job State Total Tasks lt INTEGER gt Number of tasks requested by job User lt STRING gt gt Name Name of user submitting job user Name of user submitting job job WallTime DD HH MM SS Length of time job has been S ee WallTime Limit Limit DD HH MM DD HH IMM SS DD HH MM SS__ Maximum Maximum walltime allowed to job allowed to Maximum walltime allowed to job In the above table fi
4. In this example the window is specifically for user john in group staff This information is important because processors can be reserved for particular users and groups thus causing backfill windows to be different for each person Backfill window information for a non default user group and or account can be displayed using the u g and a flags respectively A backfill window with global user group and account access can be displayed using the A flag Example 2 showbf r 16 d 3 00 00 backFill window user john group staff partition ALL Mon Feb 16 08 28 54 partition ALL 33 procs available with no time limit In this example the output verifies that a backfill window exists for jobs requiring a 3 hour runtime and at least 16 processors Specifying job duration is of value when time based access is assigned to reservations i e using SRMAXTIME Example 3 showbf m gt 128 backfill window user Jjohn group staff partition ALL Thu Jun 18 16 03 04 no procs available In this example a backfill window is requested consisting for available processors located only on nodes with over 128 MB of memory Unfortunately in the example no processors are available which meet this criteria at the present time Related Commands Use the showq command to show jobs in the various queues Use the diagnose command to show the partitions Notes See the Backfill docu
5. Maui 3 2 2 and higher restarted at a later time specifies the coefficient to be PROCWEIGHT X lt INTEGER gt 0 multiplied by a job s requested PROCWEIGHT 2500 processor count priority factor The amount of time Maui will keep a job or node record for an object no longer reported by the resource manager Useful when using a resource manager which P URGETIME 00 05 00 PURGETIME DD HH MM SS 0 drops information about a node Maui will maintain a job or node record for 5 minutes or job due to internal after the last update regarding that object received Fpa A In Sra 3 2 0 from the resource manager an higher this parameter is superseded by JOBPURGETIME and NODEPURGETIME specifies QOS specific list of delimited attributes See the flag lt ATTR gt lt VALUE gt pairs where lt ATTR gt overview for a description of ee ee maenoce 80 T000 is one of the following legal flag values 7 z QOSCFG lt QOSID gt PRIORITY FSTARGET QTWEIGHT NONE NOTE Available in Maui 3 0 6 Maui will increase the priority of jobs using QOS QTTARGET XFWEIGHT XFTARGET PLIST PDEF FLAGS or a fairness policy specification QOSFLAGS and other QOS parameters commercial and will allow up to 4 simultaneous QOS commercial jobs with up to 80 total allocated processors specifies which node features must be present on resources allocated to jobs of the QOSFEATURES 2 wide interactive
6. The amount of each job s allocation charge is directly associated with the amount of resources used i e processors by that job and the amount of time it was used for Optionally the allocation manager can also be configured to charge accounts varying amounts based on the QOS desired by the job the type of compute resources used and or the time when the resources were used both in terms of time of day and day of week The allocations manager interface provides near real time allocations management giving a great deal of flexibility and control over how available compute resources are used over the medium and long term and works hand in hand with other job management features such as Maui s throttling policies and fairshare mechanism Configuring Maui Maui interfaces with the allocations manager if the parameter BANKTYPE is specified Maui currently interfaces to QBank and RES and can also dump allocation manager interface interaction to a flat file for post processing using the type FILE Depending on the allocation manager type selected it may also be necessary to specify how to contact the allocation manager using the parameters BANKSERVER and BANKPORT When an allocations bank is enabled in this way Maui will check with the bank before starting any job For allocation tracking to work however each job must specify an account to charge or the bank must be set up to handle default accounts on a per user basis Under this confi
7. reservation staff 1 created on 4 nodes 4 tasks node003 node004 node005 node006 E ee Related Commands Use the showres command to view reservations Use the releaseres command to release reservations Use the diagnose r command to analyze and present detailed information about reservations Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved ___ setspri setspri PRIORITY r JOB Purpose Set or remove and absolute or relative system priority on a specified job Permissions This command can be run by any Maui Scheduler Administrator Parameters JOB Name of job PRIORITY System priority level By default this priority is an absolute priority overriding the policy generated priority value Range is 0 to clear 1 for lowest 1000 for highest If the r flag is specified the system priority is relative adding or subtracting the specified value from the policy generated priority If a relative priority is specified any value in the range 1000000000 is acceptable Flags h Help for this command r Set relative system priority on job Description This command allows you to set or remove a system priority level for a specified job Any job with a system priority level set is guaranteed a higher priority than jobs without a system priority Jobs with higher system priority settings have priority over jobs with lower system priority settings Example 1 setspri
8. 7 job depends on another job reaching a certain state rejected by policy job start is prevented by a throttling policy If a job cannot run on a particular node one of the following per node reasons will be given Class Node does not allow required job class queue CPU Node does not possess required processors Disk Node does not possess required local disk Features Node does not possess required node features Memory Node does not possess required real memory Network Node does not possess required network interface State Node is not Idle or Running The checkjob command displays the following job attributes Attribute Walue Value Description a ae lt sTRIN G gt E of account associated with Length of time job actually ran Actual Run Time DD HH MM SS NOTE This info only display in simulation mode Arch STRING gt Node architecture required by job lt CLASS NAME gt lt CLASS Name of class queue required by COUNT gt job and number of class initiators required per task eee Resources Per Task acres Amount of local disk required by Disk lt INTEGER gt i in MB Exec Size lt INTEGER gt Size of job executable in MB Executable STRING gt Name of job executable ae bracket delimited list of Features lt STRINGSs Node features required by job Group lt STRING gt Ez of UNIX group associated with job
9. Idle Job Limits Constrains the total cumulative resources available to idle jobs at a given time System Job Limits Constrains the maximum resource requirements of any single job These limits can be applied to any job credential user group account QOS and class or on a system wide basis Additionally QoS s may be configured to allow limit overrides to any particular policy For a job to run it must meet all policy limits Limits are applied using the CFG set of parameters particularly USERCFG GROUPCFG ACCOUNTCFG QOSCFG CLASSCEKG and SYSTEMCFG Limits are specified by associating the desired limit to the individual or default object The usage limits currently supported by Maui listed in the table below INAME UNITS DESCRIPTION EXAMPLE Limits the number of jobs a credential may MAXJOB _ of jobs have active Starting IMAXJOB 8 or Running at any given time Limits the total number of dedicated processors which can be allocated by active jobs at any given time MAXPROC of processors MAXPROC 32 Limits the number of outstanding processor seconds a credential may have allocated at any given time For example if a user has a 4 processor job which will complete in 1 hour and a 2 processor job which will complete in 6 lt of processors gt hours he has 4 1 MAXPS lt walltime gt 3600 2 6 3600 16 3600 outstanding processor seconds The outstanding processo
10. If set to SHARED allows a standing reservation to utilize resources already allocated to SRACCESS 2 SHARED other non job reservations Otherwise these other reservations will block resource access See Managing Reservations Standing reservation 2 may access resources allocated to existing standing and administrative reservations specifies that jobs with the associated accounts may use the _ resources contained within this Gobs using the account ops or staff are granted reservation access to the resources in standing reservation 1 SRACCOUNTLIST 1 ops staff specifies the account to which SRCHARGEACCOUNT 1 steve maui will charge all idle cycles within the reservation via the Maui will charge all idle cycles within reservations allocation bank supporting standing reservation 1 to user steve SRCFG fast STARTTIME 9 00 00 ENDTIME 15 00 00 specifies attributes of a standing SRCES rae ise ee reservation Available in Maui 3 2 and higher See Managing Maui will create a standing reservation running from Reservations for details 9 00 AM to 3 00 PM on nodes 1 through 4 accessible by jobs with QOS high or low specifies that jobs requiring any of these classes may use the SRCLASSLIST 2 interactive SRCLASSLIST X list of valid class names NONE Nap ES maui will allow all jobs requiring any of the class
11. email users regarding statistics of all completed jobs email users only when certain criteria are met ie Dear John you submitted job X requesting 128MB of memory per task It actually utilized 253 MB of memory per task potentially wreaking havoc with the entire system Please improve your resource usage estimates in future jobs update system databases take system actions based on job completion statistics NOTE some of these fields may be set to zero if the underlying OS Resource Manager does not support the necessary data collection Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ Appendix A Case Studies A 1 Case 1 Mixed Parallel Serial Heterogeneous Cluster A 2 Case 2 Partitioned Timesharing Cluster A 3 Case 3 Development O2K A 4 Case 4 Standard Production SP2 A 5 Case 5 Multi Queue Cluster with QOS and Charge Rates Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ A 1 Case Study Mixed Parallel Serial Homogeneous Cluster Overview A multi user site wishes to control the distribution of compute cycles while minimizing job turnaround time and maximizing overall system utilization Resources Compute Nodes 64 2 way SMP Linux based nodes each with 512 MB of RAM and 16 GB local scratch space Resource Manager OpenPBS 2 3 Network 100 MB switched ethernet Workload Job Size range in size from 1 to 32 processors
12. CPU efficiency of job 10 Ooo oo oo oo COO OO OC OD OC 0 0 0 0 0 0 User dsheppar dsheppar dsheppar dsheppar ebylaska kossi xztang moorejt mukho zhong wengel vertex kossi jimenez vertex vertex wengel kudo vertex yshi jshoemak rich0d01 rampi Efficiency Group daf daf daf daf dnavy daf daf daf dnavy govt univ univ daf dnavy univ univ univ daf univ univ daf daf univ 98 82 Nodes H gi hb NM OOD AOrRrA RARER E w Remaining G G A E n O O ie 10 18 T9 T93 20 20 Zils 233 23 26 26 RESES IE i504 203 0 33 ems 272 223 241 242 250 524 2 30 273 54 04 10 24 26 18 37 Sli 10 Oks 50 51 10 49 58 05 42 47 29 44 732 30 51 53 30 30 45 27 21 17 51 25 41 Fri Fri Fri Fri Fri Fri Thu Fri Thu Fri Thu Fri Fri Fri Fri Fri Fri Fri Fri Fri Fri Fri Fri Aug Aug Aug Aug Aug Aug Aug Aug 29 7 29 29 29 7 29 7 29 28 1 29 28 1 29 28 29 29 29 ye alk 29T 2901 291 29 1 29 i 29 1 29 Current expansion factor of job where XFactor QueueTime WallClockLimit WallClockLimit Quality Of Service specified for job User owning job Primary group of job owner Number of processors being used by the job Remaining Time the job has until it has reached its wall clock limit Time specified in HH MM SS notation StartTime
13. Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved L 16 4 Simulation Specific Configuration Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 17 0 Miscellaneous Features RESDEPTH Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 17 1 Feedback Script The Feedback Script facility allows a site to provide job performance information to users at job completion time When a job completes the program pointed to by the FEEDBACKPROGRAM parameter is called with a number of command line arguments The site is responsible for creating a program capable of processing and acting upon the contents of the command line The command line arguments passed are a follows job name user name user email final job state QOS requested epoch time job was submitted epoch time job started epoch time job completed job XFactor job wallclock limit processors requested memory requested average per task cpu load maximum per task cpu load average per task memory usage maximum per task memory usage For many sites the feedback script is useful as a means of letting user s know that accuracy of their wallclock limit estimate as well as the cpu efficiency and memory usage pattern of their job The feedback script may be used as a mechanism to do any of the following
14. Job Completion 7 lt STRING gt Completed One of Completed Removed NotRun i Class queue required by job specified as square lt STRING gt DEFAULT 1 bracket list of lt QUEUE gt lt QUEUE INSTANCE gt requirements ie batch 1 _ 9 lt INTEGER gt 0 Epoch time when job was submitted Dispatch 10 lt INTEGER gt 0 Epoch time when scheduler requested job begin Time executing Start Time 11 lt INTEGER gt 0 Epoch time when job began executing NOTE usually identical to Dispatch Time oo 12 lt INTEGER gt 0 Epoch time when job completed execution lt STRING gt NONE Name of required network adapter if specified lt STRING gt NONE Required node architecture if specified lt STRING gt NONE Required node operating system if specified at Comparison for determining compliance with one of gt gt lt lt gt required node memory Amount of required configured RAM in MB on 17 lt INTEGER gt 0 each node Comparison for determining compliance with required node disk 18 one of gt gt lt lt gt Comparison Required 9 lt INTEGER gt 0 Amount of required configured local disk in MB on Node Disk each node square bracket enclosed list of node features required pone 20 So TRING NONE by job if specified ie fast ethernet lt INTEGER gt 0 Epoch time when job met all fairness po
15. MAXJOB Maui 3 0 7 and higher This policy constrains the number of total independent jobs a given node may run simultaneously It can only be specified via the NODECFG parameter MAXJOBPERUSER Maui 3 0 7 and higher This policy constrains the number of total independent jobs a given node may run simultaneously associated with any single user Like MAXJOB it can only be specified via the NODECFG parameter MAXLOAD Maui 3 2 2 and higher MAXLOAD constrains the CPU load the node will support as opposed to the number of jobs If the node s load exceeds the MAXLOAD limit and the NODELOADPOLICY parameter is set to ADJUSTSTATE the node will be marked busy Under Maui 3 0 the max load policy could be applied system wide using the parameter NODEMAXLOAD Node policies are used strictly as constraints If a node is defined as having a single processor or the NODEACCESSPOLICY is set to DEDICATED and a MAXJOB policy of 3 is specified Maui will probably not run more than one job per node A node s configured processors must be specified so that multiple jobs may run and then the MAXJOB policy will be effective The number of configured processors per node is specified on a resource manager specific basis PBS for example allows this to be adjusted by setting the number of virtual processors np per node in the PBS nodes file Example maui cfg NODECFG node024 MAXJOB 4 MAXJOBPERUSER 2 NODECFG node025 MAXJOB 2 NODECFG node026
16. NONE Sample Workload Trace SP02 2343 0 20 20 570 519 86400 Removed batch 1 887343658 889585185 889585185 889585411 ethernet R6000 AIX43 gt 256 gt 0 NONE 889584538 20 0 0 2 0 test cmd 1001 6 678 08 0 1 0 0 0 0 O NONE 0 NONE NONE NONE NONE NONE 16 3 2 Creating New Workload Traces Because workload traces and workload statistics utilize the same format there are trace fields which provide information that is valuable to a statstical analysis of historical system performance but not necessary for the execution of a simulation i Particularly in the area of time based fields there exists an opportunity to overspecify Which time based fields are important depend on the setting the the JOBSUBMISSIONPOLICY parameter JOBSUBMISSIONPOLICY Value Critical Time Based Fields WallClock Limit Submission Time StartTime Completion Time NORMAL CONSTANTJOBDEPTH CONSTANTPSDEPTH Completion Time NOTE 1 Dispatch Time should always be identical to Start Time NOTE 2 In all cases the difference of Completion Time Start Time is used to determine actual job run time NOTE 3 System Queue Time and Proc Seconds Utilized are only used for statistics gathering purposes and will not alter the behavior of the simulation NOTE 4 In all cases relative time values are important i e Start Time must be greater than or equal to Submission Time and less than Completion Time
17. Task List The list of nodes on which the task instances have been located Req Statistics Statistics tracking resource utilization 3 2 1 2 Nodes As far as Maui is concerned a node is a collection of resources with a particular set of associated attributes In most cases it fits nicely with the canonical world view of a node such as a PC cluster node or an SP node In these cases a node is defined as one or more CPU s memory and possibly other compute resources such as local disk swap network adapters software licenses etc Additionally this node will described by various attributes such as an architecture type or operating system Nodes range in size from small uniprocessor PC s to large SMP systems where a single node may consist of hundreds of CPU s and massive amounts of memory Information about nodes is provided to the scheduler chiefly by the resource manager Attributes include node state configured and available resources 1 e processors memory swap etc run classes supported etc 3 2 1 3 Advance Reservations An advance reservation is an object which dedicates a block of specific resources for a particular use Each reservation consists of a list of resources an access control list and a time range for which this access control list will be enforced The reservation prevents the listed resources from being used in a way not described by the access control list during the time range specified For example
18. gt 512 MB required over a day to complete It looks like there is nothing that can start right now and we will have to live with four idle nodes Let s look at the reservation which is blocking the start of our single processor jobs gt showres This command shows all reservations currently on the system Notice that all running jobs have a reservation in place Also there is one reservation for an idle job Indicated by the T in the S or State column This is the reservation that is blocking our serial jobs This reservation was actually created by the backfill scheduler for the highest priority idle job as a way to prevent starvation while lower priority jobs were being backfilled The backfill documentation describes the mechanics of the backfill scheduling more fully Let s see which nodes are part of the idle job reservation gt showres n fr8n01 963 0 All of our four idle nodes are included in this reservation It appears that everything is functioning properly Let s step further forward in time gt schedctl s 100I gt showstats v We now know that the scheduler is scheduling efficiently So far system utilization as reported by showstats v looks very good One of the next questions is is it scheduling fairly This is a very subjective question Let s look at the user and group stats to see if there are any glaring problems gt showstat
19. lt INTEGER gt 0 time priority weight QOSQTWEIGHT 5 QOSWEIGHT 3 10 specifies the expansion factor QOSXFTARGET 3 5 0 QOSXFTARGET X lt DOUBLE gt NONE target used ina job s Target jobs requesting a QOS of 3 will have their priority Factor priority calculation grow exponentially as the job s minimum expansion factor approaches 5 0 D 7 XFWEIGHT 0 100 Mere A which will QoSxFWEIGHT 2 1000 e added to the base QOSXFWEIGHTIX lt INTEGER gt 0 XFWEIGHT for all jobs using Gobs using QOS 2 will have a XFWEIGHT of 1100 QOS X while jobs using other QOS s will have an XFWEIGHT of 100 QUEUETIMECAP 0 10000 specifies the maximum allowed QUEUETIMEWEIGHT 0 10 QUEUETIMECAP X lt DOUBLE gt 0 NO CAP pre weighted queuetime priority a job that has been queued for 40 minutes will have its factor queuetime priority factor calculated as Priority QUEUETIMEWEIGHT MIN 10000 40 specifies multiplier applied to a job s queue time in minutes to QUEUETIMEWEIGHT 0 20 QUEUETIMEWEIGHT X lt INTEGER gt 1 determine the job s queuetime a job that has been queued for 4 20 00 will have a priority factor queuetime priority factor of 20 260 one of the following specifies who can create admin RESCTLPOLICY ANY RESCTLPOLICY ADMINONLY
20. ANY ADMINONLY reservations Available in Maui any valid user can create an arbitrary admin 3 2 and higher reservation specifies the maximum number of reservations which can be on any single node IMPORTANT NOTE on large way SMP RESDEPTH lt INTEGER gt 24 ystems dus valve pien aasi Minara 64 be increased To be on the safe side this value should be approximately twice the average sum of admin standing and job reservations present p ae RESERVATIONDEPTH 0 4 specifies how many priority RESERVATIONOOSLIST 0 1 3 5 RESERVATIONDEPTH X lt INTEGER gt 1 reservations are allowed in the associated reservation stack jobs with QOS values of 1 3 or 5 can have a cumulative total of up to 4 priority reservations p RESERVATIONPOLICY CURRENTHIGHEST one of the following specifies how Maui reservations ppspRVATIONDEPTH 2 RESERVATIONPOLICY CURRENTHIGHEST HIGHEST CURRENTHIGHEST will be handled See also NEVER RESERVATIONDEPTH Maui will maintain reservations for only the two currently highest priority jobs RESERVATIONDEPTH 0 4 specifies which QOS levels have RESERVAT IONQOSLIST 0 1 3 5 RESERVATIONQOSLIST X one or more QOS values or ALL ALL access to the associated reservation stack jobs with QOS values of 1 3 or 5 can have a cumulative total of up to 4 priority reservations Period of time Maui will continue to attempt to start a job RESERVATIONRETRYTIME X DD HH MM SS 0 in a reservation when job start failures are detecte
21. RAM installed Let s verify that the idle nodes do not have enough memory configured gt diagnose n grep e Idle e Name The grep gets the command header and the Idle nodes listed All idle nodes have only 256 MB of memory installed and cannot be allocated to this job The diagnose command can be used with various flags to obtain detailed information about jobs nodes reservations policies partitions etc The command also performs a number of sanity checks on the data provided and will present warning messages if discrepancies are detected Let s see if the other single processor jobs cannot run for the same reason gt diagnose j grep Idle grep 1 The grep above selects single processor Idle jobs The 14th indicates that most single processor jobs currently in the queue require gt 256 MB of RAM but a few do not Let s examine job fr8n01 1154 0 gt checkjob fr8n01 1154 0 The rejection reasons for this job indicate that the four idle processors cannot be used due to ReserveTime This indicates that the processors are idle but that they have a reservation in place that will start before the job being checked could complete Let s look at one of the nodes gt checknode fr10n09 The output of this command shows that while the node is idle it has a reservation in place that will start in a little over 23 hours All idle jobs which did not require
22. SERVERHOST This specifies where Maui will run It allows Maui client commands to locate the Maui server It must specify the fully qualified hostname of the machine on which Maui will run Example SERVERHOST cw psu edu SERVERPORT This specifies the port on which the Maui server will listen for client connections Unless the default port of 40559 is unacceptable this parameter need not be set Example SERVERPORT 50001 ADMIN1 Maui has 3 major levels of admin access Users which are to be granted full control of all Maui functions should be indicated by setting the ADMIN1 parameter The first user in this list is considered the primary admin It is the ID under which Maui should always run Maui will only run under the primary admin user id and will shut itself down otherwise In order for Maui to properly interact with both PBS and Loadleveler it is important that the primary Maui admin also be configured as a resource manager admin within each of those systems Example ADMIN1 joe charles RMTYPE X Maui must be told which resource manager s to talk to Maui currently has interfaces to Loadleveler Wiki and PBS To specify a resource manager typically only the resource manager type needs to be indicated using the keywords LL WIKI or PBS Example RMTYPE 0O PBS The array index in the parameter name allows more than one resource manager to be specified In these multiple resource manager situations additional parameter
23. Then what s this SRDEPTH parameter for This parameter remedies a situation which might occur when a job is submitted and cannot run immediately because the system is completely backlogged with jobs In such a case available resources may not exist for two days out and Maui will reserve them for this job When midnight arrives Maui attempts to roll its standing reservations but here a problem arises This job has now allocated the resources needed for the standing reservation two days out Maui cannot reserve the resources for the standing reservation because they are already claimed by the job The standing reservation reserves what it can but it is now smaller than it should be or possibly even empty If a standing reservation is smaller than it should be Maui will attempt to add resources every iteration until it is fully populated However in the case of this job it is not going to let go of the resources it has and the standing reservation is out of luck The SRDEPTH parameter allows a site to create standing reservations deep into the future allowing them to claim the resources first and preventing this problem If partial standing reservations are detected on a system it may be an indication that the SRDEPTH parameter should be increased In the example above the SRPERIOD parameter is set to INFINITY With this setting a single permanent standing reservation is created and the issues of resource contention do not exist While this
24. These include RMNAME RMPORT RMHOST RMTIMEOUT RMAUTHTYPE RMCONFIGFILE and RMNMPORT The RMNAME parameter allows a site to associate a name with a particular resource manager so as to simplify tracking of this interface within Maui To date most sites have chosen to setup only one resource manager per scheduler making this parameter largely unnecessary RMHOST and RMPORT allow specification of where the resource manager is located These parameters need only to be specified for resource managers using the WIKI interface or with PBS when communication with a non default server is required In all other cases the resource manager is automatically located The maximum amount of time Maui will wait on a resource manager call can be controlled by the RMTIMEOUT parameter which defaults to 30 seconds Only rarely will this parameter need to be changed RMAUTHTYPE allows specification of how security over the scheduler resource manager interface is to be handled Currently only the WIKI interface is affected by this parameter The allowed values are documented in the RMAUTHTYPE parameter description Another RM specific parameter is RMCONFIGFILE which specifies the location of the resource manager s primary config file and is used when detailed resource manager information not available via the scheduling interface is required It is currently only used with the Loadleveler interface and needs to only be specified when using Maui meta scheduling cap
25. be rolled see Logging Overview LOGFILEMAXSIZE 50000000 Log files will be rolled when they reach 50 MB in size LOGFILEROLLDEPTH LOGLEVEL MAXJOBPERUSER lt INTEGER gt lt INTEGER gt 0 9 lt INTEGER gt lt INTEGER gt 0 No Limit number of old log files to maintain i e when full maui log will be renamed maui log 1 maui log 1 will be renamed maui log 2 NOTE Only available in Maui 3 0 5 and higher see Logging Overview specifies the verbosity of Maui logging where 9 is the most verbose NOTE each logging level is approximately an order of magnitude more verbose than the previous level see Logging Overview maximum number of active jobs allowed at any given time See note for Maui 3 0 versions MAXJOBQUEUEDPERUSER lt INTEGER gt lt INTEGER gt 0 No Limit maximum number of idle jobs which can be considered for scheduling and which can acquire system queue time for increasing job priority See note for Maui 3 0 versions MAXNODEPERUSER MAXPEPERUSER MAXPROCPERUSER MAXPSPERUSER MAXWCPERUSER MEMWEIGHT X lt INTEGERS gt lt INTEGER gt lt INTEGERS gt lt INTEGER gt lt INTEGER gt lt INTEGER gt lt INTEGERS gt lt INTEGER gt DD HH MM SS DD HH MM SS lt INTEGER gt 0 No Limit 0 No Limit 0 No Limit 0 No Limit 0 No Limit max
26. between SRTASKCOUNT and the hosts in SRHOSTLIST does not necessarily mean that Maui will place one task on each host If for example nodeO11 and node012 were 2 processor SMP nodes with 1 GB of memory Maui could locate 2 tasks on each of these nodes leaving only two more tasks to be placed Maui will place tasks on nodes according to the policy specified with the NODEALLOCATIONPOLICY parameter If the hostlist provides more resources than what is required by the reservation as specified via SRTASKCOUNT Maui will simply select the needed resources within the set of nodes listed If SRHOSTLIST is specified but SRTASKCOUNT is not Maui will pack as many tasks as it can onto ALL of the listed nodes For example SRNAME 1 debug SRHOSTLIST 1 node001 node002 node003 node004 SRUSERLIST 1 helpdesk SRGROUPLIST 1 operations sysadmin SRPERIOD 1 INFINITY This standing reservation appears much simpler Since SRRESOURCES is not specified it will allocate all processors on each of the nodes listed in SRHOSTLIST Since a start and end time are not specified the reservation will be in force all day long Since SRDAYS is not specified the reservation will be enabled every day of the week Ok here come a couple of curve balls First note that standing reservation 1 debug has two access parameters set SRUSERLIST and SRGROUPLIST Reservations can be accessed by any one of the access lists specified In this case either user
27. credentials history allocated resources and resource utilization diagnose j display summarized job information and any unexpected state feigos sevmodify QoS ofexistingjob O Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved ___ 4 4 Reservation Management Commands Maui exclusively controls and manages all advance reservation features including both standing and administrative reservations The table below covers the available reservation management commands The Command Overview lists all available commands Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 4 5 Policy Config Management Commands Maui allows dynamic modification of most scheduling parameters allowing new scheduling policies algorithms constraints and permissions to be set at any time Changes made via Maui client commands are temporary and will be overridden by values specified in Maui config files the next time Maui is shutdown and restarted The table below covers the available configuration management commands The Command Overview lists all available commands Command Flags Description changeparam immediately change parameter value EE control scheduling behavior 1 e stop start scheduling recycle _ shutdown etc showconfig display settings of all configuration parameters Copyright 2000 2002 Supercluster Research and
28. jobs with a QOS value of 2 may only run on nodes QOSFEATURES X one or more node feature values or ANY ANY neat This parameter with the feature wide AND the feature interactive takes a name as an array cot index one or more of the following space delimited IGNJOBPERUSER IGNPROCPERUSER IGNNODEPERUSER IGNPSPERUSER IGNJOBQUEUEDPERUSER IGNJOBPERGROUP a eae Ha specifies the attributes of the IGNJOBQUEUEDPERGROUP pie saci a See QOSFLAGS 1 ADVRES IGNMAXJOBPERUSER IGNJOBPERACCOUNT i 2 jobs with a OOS value of 1 d QOSFLAGS X IGNPROCPERACCOUNT NONE Overview section for details jobs with a QOS value of 1 must run in an advance IGNPSPERACCOUNT 3 NOTE some flags are only reservation and can ignore the MAXJOBPERUSER IGNJOBQUEUEDPER ACCOUNT supported under Maui 3 1 and POlicy IGNSYSMAXPROC IGNSYSMAXTIME later IGNSYSMAXPS IGNSRMAXTIME IGNUSER IGNGROUP IGNACCOUNT IGNSYSTEM IGNALL PREEMPT DEDICATED RESERVEALWAYS USERESERVED NOBF NORESERVATION RESTARTPREEMPT specifies the priority associated goSPRIORITY 2 1000 QOSPRIORITY X lt INTEGER gt 0 with this QOS NOTE only used in Maui 3 0 x set the priority of QOS 2 to 1000 specifies the target job QOSQTTARGET X DD HH MM SS NONE queuetime associated with this QOSQTTARGET 2 00 00 QOS specifies the per QOS queue p QOSQTWEIGHT X
29. that QOS is listed in the system default configuration QDEF or QLIST or if the QOS is specified in the QDEF or QLIST of a user group account or class associated with that job The diagnose Q command can be used to obtain information about the current QOS configuration 7 3 2 QoS Enabled Privileges The privileges enabled via QoS settings may be broken into one of the following categories Special Prioritization Service Access and Constraints Override Policies and Policy Exemptions All privileges are managed via the QOSCFG parameter 7 3 2 1 Special Prioritization IFSTARGET PRIORITY Assign priority to all jobs requesting particular QoS QTTARGET QTWEIGHT XFTARGET XFWEIGHT Example QOSCFG geo PRIORITY 10000 7 3 2 2 Service Access and Constraints The QoS facility can ne used to enable special service and or disable default services All services are enabled disabled by setting the QoS FLAG attribute Flag Name Description jobs should not share compute resources with any other job These jobs will only run on nodes which are idle and will not allow other jobs to use resources on allocated nodes even if additional resources are available INOBF j ob cannot be considered for backfilled NORESERV ATION p should never reserve resources regardless of priority job may be preempted by higher priority PREEMPTEE ee jobs PREEMPT OR preempt lower priority PREEMPTEE RESERVEALWAYS ob should create resource res
30. users can only place holds on their jobs Jobs with a user hold in place will have a Maui state of Hold or UserHold depending on the resource manager being used System Holds The second category of hold is the system hold This hold is put in place by a system administrator either manually or by way of an automated tool As with all holds the job is not allowed to run so long as this hold is in place A batch administrator can place and release system holds on any job regardless of job ownership However unlike a user hold a normal user cannot release a system hold even on his own jobs System holds are often used during system maintenance and to prevent particular jobs from running in accordance with current system needs Jobs with a system hold in place will have a Maui state of Hold or SystemHold depending on the resource manager being used Batch Holds Batch holds constitute the third category of job holds These holds are placed on a job by the scheduler itself when it determines that a job cannot run The reasons for this vary but can be displayed by issuing the checkjob lt JOBID gt command Some of the possible reasons are listed below No Resources the job requests resources of a type or amount that do not exist on the system System Limits the job is larger or longer than what is allowed by the specified system policies Bank Failure the allocations bank is experiencing failures No Allocations the job requests use o
31. you may wish to dive in and start the scheduler in NORMAL mode This admin manual and the accompanying man pages should introduce you to the relevant issues and commands To start the scheduler in NORMAL mode take the following step gt maui That should be all that is needed to get you started Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 3 0 Basic Maui Overview 3 1 Layout of Maui Components 3 2 Scheduling Environments and Objects 33 Job Flow ail 3 4 Configuring the Scheduler Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 3 1 File Layout Maui is initially unpacked into a simple one deep directory structure as shown below Note that some of the files i e log and statistics files will be created as Maui is run MAUIHOMEDIR ___ maui cfg general config file containing information required by both the Maui server and user interface clients ___ smaui private cfg config file containing private information required by the Maui server only ___fs cfg fairshare config file used in Maui 3 0 6 and earlier l maui ck Maui checkpoint file I maui pid Maui lock file to prevent multiple instances I_ log directory for Maui log files REQUIRED BY DEFAULT _ s maui log Maui log file I__ maui log 1 previous rolled Maui log file ___ stats directory for Maui statistics files REQUIRED BY DEFAULT l_
32. 10 fr13n03 24 0 Job System Priority Adjusted In this example a system priority of 10 is set for job fr13n03 24 0 Example 2 setspri 0 fr1i3n03 24 0 Job System Priority Adjusted In this example system priority is cleared for job fr13n03 24 0 Example 3 gt setspri r 100000 job 00001 Job System Priority Adjusted In this example the job s priority will be increased by 100000 over the value determine by configured priority policy Related Commands Use the check job command to check the system priority level if any for a given job Notes None Copyright 1998 Maui High Performance Computing Center All rights reserved Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ showbf showbf A show information accessible by A ny user group or account a ACCOUNT c CLASS d DURATION f FEATURELIST g GROUP h m MEMCMP MEMORY n NODECOUNT p PARTITION q QOS u USER v VERBOSE Purpose Shows what resources are available for immediate use NOTE if specific information is not specified showbf will return information for the user and group running but with global access for other fields For example if q lt QOS gt is not specified Maui will return backfill information for a job which could magically access all QOS based resources ie resources covered by reservations with a QOS based ACL if c lt CLASS gt is no
33. 21102 1 4 20906 1 4 20604 e 20180 1 6 20024 175 19916 dea 6 19097 1 2 12547 1 0 9390 10 6434 Node Hours Name of job Calculated job priority User ozturan jason jason moraiti moraiti kdeacon jpar jpar chori moorej moorejt K ebylaska dsheppar zhong jacob Group govt asp asp univ univ pdc dnavy dnavy univ daf daf dnavy daf govt univ 25 33 Hours Nodes 64 128 128 128 128 64 WCLimit 0 16 40 2 23 00 00 2 0 220 00 00 00 200 259 2554 259 200 59 00 00 59 00 00 00 Class batch medium medium batch batch batch batch batch batch batch batch Thu Fri Fri Fri Fri Fri Fri Fri Fri Fri Fri Fri Aug Aug Aug Aug Aug Aug Aug Aug Aug Aug Aug Aug Aug Aug Aug Current expansion factor of job where XFactor QueueTime WallClockLimit WallClockLimit Quality Of Service specified for job User owning job Primary group of job owner Minimum number of processors required to run job Wall clock limit specified for job Time specified in HH MM SS notation Class requested by job After displaying the job listing the command summarizes the workload in the idle queue and indicates the total workload backlog in node hours The value in parenthesis indicates the
34. 31 frl7n06 1517 0 cholik Idle 16 24 00 00 Fri Aug 29 06 45 46 fr28n13 706 0 moorejt Idle 16 5 55 00 Fri Aug 29 10 53 53 fr17n16 1550 0 moorejt Idle 16 7 55 00 Fri Aug 29 10 53 54 frl7n12 1528 0 ebylaska Idle 16 3 59 59 Eri A g 29 12511330 fr28n15 4356 0 dsheppar Idle 16 3 00 00 Fri Aug 29 14 01 42 fr28n09 50 0 dsheppar Idle 16 3 00 00 Fri Aug 29 14 01 59 fr28n09 51 0 zhong Idle 8 13 55 00 Fri Aug 29 14 07 16 fr17n16 1551 0 jacob Idle 4 4 00 00 Fri Aug 29 12 51 19 15 Idle Job s NON QUEUED JOBS JOBNAME USERNAME STATE PROC CPULIMIT QUEUETIME frl7n02 1476 0 vertex Idle 1 22 00 00 Thu Aug 28 23 48 16 fr17n05 1392 0 vertex SystemHold al 22 00 00 Thu Aug 28 23 49 51 frl7n10 1449 0 vertex Idle 1 22 00 00 Tue Aug 26 23 49 51 fr28n03 1674 0 maxia UserHold 8 23 56 00 Mon Aug 25 16 22 10 f r28n05 1581 0 sidt UserHold I 1 00 00 Sun Jul 27 12 46 17 fr28n05 2092 0 vertex Idle 1 22 00 00 Thu Aug 28 23 48 40 fr28n13 705 2 gigi NotQueued 32 15 58 00 Fri Aug 29 10 49 01 fr28n13 705 3 gigi NotQueued 32 13 58 00 Fri Aug 29 10 49 01 fr17n08 1349 7 dsheppar BatchHold al 2 00 00 Fri Aug 29 13 34 44 fr28n15 4355 1 dsheppar Idle 1 2 00 00 Fri Aug 29 13 35 04 fr28n15 4355 2 dsheppar Deferred I 2 00 00 Fri Aug 29 13 35 04 fr28n15 4355 3 dsheppar Idle 2 00 00 Fri Aug 29 13 35 04 Total Jobs 63 Active Jobs 36 Idle Jobs 15 on Queued Jobs 12 The output of this command is divided into three parts Active Jobs Idle Jobs and Non Queued Jobs Active jobs are those tha
35. ACL is satisfied by any object not explicitly listed by a NOT entry Also if an object matches a NOT entry the associated job is excluded from the reservation even if it meets other ACL requirements For example a QOS 3 job requesting account projectX will be denied access to the reservation even though the job QOS matches the QOS ACL Note that the ability to specify NOT ACLs is only enabled in Maui 3 0 7 and higher 7 1 5 2 5 Resource Allocation Behavior As mentioned above standing reservations can operate in one of two modes floating or non floating essentially node locked A floating reservation is created when a SRTASKCOUNT is specified and SRHOSTLIST is either not specified or specified with more resources than are needed to fulfill the SRTASKCOUNT requirement If a reservation is non floating Maui will allocate all resources specified by the SRHOSTLIST parameter regardless of node state job load or even the presence of other standing reservations Maui interprets the request for a non floating reservation as stating I want a reservation on these exact nodes no matter what If a reservation is configured to be floating Maui takes a more relaxed stand searching through all possible nodes to find resources meeting standing reservation constraints Only Idle Running or Busy node will be considered and further only considered if no reservation conflict is detected The parameter SRACCESS can be used to modify this
36. Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 15 6 Collecting Performance Information on Individual Jobs Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 16 0 Simulations 16 1 Simulation Overview 16 2 Resource Traces 16 3 Workload Traces 16 4 Simulation Specific Configuration Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved L 16 1 Simulation Overview 16 1 1 Test Drive If you want to see what the scheduler is capable of the simulated test drive is probably your best bet This allows you to safely play with arbitrary configurations and issue otherwise dangerous commands without fear of losing your job In order to runa simulation you need a simulated machine defined by a resource trace file and a simulated set of jobs defined by a workload trace file Rather than discussing the advantages of this approach in gory detail up front let s just get started and discuss things along the way Issue the following commands gt vi maui cfg change SERVERMODE NORMAL to SERVERMODE SIMULATION add SIMRESOURCETRACEFILE _ traces Resource Tracel1 add SIMWORKLOADTRACEFILE traces Workload Tracel1 add SIMSTOPITERATION 1 the steps above specified that the scheduler should do the following 1 Run in Simulation mode rather than in Normal or live mode
37. Development Group All Rights Reserved 4 6 End User Commands While the majority of Maui commands are tailored for use by system administrators a number of commands are designed to extend the knowledge and capabilities of end users The table below covers the commands available to end users The Command Overview lists all available commands Command Flags Description cancel ob cancel existing job ebasiint display job state resource requirements environment constraints credentials history allocated resources and resource utilization showbf show resource availability for jobs with specific resource requirements sho wq display detailed prioritized list of active and idle jobs showstart show estimated start time of idle jobs estimated start time of idle show estimated start time of idle jobs pron detailed usage statistics for users groups and accounts which the showstats end user has access to Copyright 2000 2002 Supercluster Research and Development Group All Rights E HMMM EMI 1 iii 4 7 Miscellaneous Commands The table below covers a number of additional commands which do not fully fit in prior categories The Command Overview lists all available commands Command Command Flags Description eeses Ss resetstats reset internal statistics Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 5 0 Assigning
38. Down or Drained Eventi ppe e DEFINED or NONE respectively DRAINED NOTE node state can be modified using the nodectl command Event Time 3 lt EPOCHTIME gt 1 Ume eventoceuried Leurrentiy ignored for COMPUTENODE Resource ID 4 lt STRING gt N A resources this should be the name of the node 5 lt STRING gt NONE name of resource manager resource is associated with aa amount of virtual memory in Swap P KINTEGER gt i MB configured on node amount of real memory in 7 lt INTEGER gt 1 MB configured on node i e RAM Configured amount of local disk in MB ne KINTEGER gt l on node available to batch jobs Configured 9 lt INTEGER gt 1 number of processors Processors configured on node number of frame containing 10 lt INTEGE node SP2 only Number of first frame slot used by node SP2 only lt THE an TEGER gt esmen G gt NONE node operating system lt STRING gt NONE hode architecture i i square bracket delimited list of 15 lt STRING gt NONE node features attributes ie amd s1200 Cont aned square bracket delimited list of 16 lt STRING gt batch 1 CLASSNAME CLASSCOUNT Run Classes pairs i i b p Number of frame slots used by node SP2 only square bracket delimited list of 7 lt STRING gt NONE configured network adapters ie atm fddi ethernet 8 avoui 1 9 lt STRING gt NONE NONE 0 lt STRING gt NONE NONE 1
39. E Fairshare Target NONE Current Fairshare Usage 20 0 Priority Weights FSWEIGHT 100 FSUSERWEIGHT 10 FSGROUPWEIGHT 20 FSACCOUNTWEIGHT 30 FSQOSWEIGHT 40 FSCLASSWEIGHT 0 In this example the Fairshare component calculation would be as follows Priority 100 10 5 20 0 30 10 40 O0 0 0 User A is 5 below his target so fairshare increases the total fairshare factor accordingly Group B has no target so group fairshare usage is ignored Account C is above its 10 above its fairshare usage target so this component decreases the job s total fairshare factor QOS 3 is 15 over its target but the in the target specification indicates that this is a floor target only influencing priority when fairshare usage drops below the target value Thus the QOS 3 fairshare usage delta does not influence the fairshare factor Fairshare is a great mechanism for influencing job turnaround time via priority to favor a particular distribution of jobs However it is important to realize that fairshare can only favor a particular distribution of jobs it cannot force it If user X has a fairshare target of 50 of the machine but does not submit enough jobs no amount of priority favoring will get user X s usage up to 50 See the Fairshare Overview for more information 5 1 2 3 Resource RES Component Weighting jobs by the amount of resources requested allows a site to favor particular types of jobs Such prioritization
40. Efficiency Average job efficiency UtilizedTime DedicatedTime Est Backlog Estimated backlog of queued work in hours Avg Backlog Average backlog of queued work in hours Example 5 o showstats u User Statistics Initialized Tue Aug 26 14 32 39 Running Completed UserName UID Jobs Procs ProcHours Jobs PHReq PHDed FSTgt AvgXF MaxXF AvgQH Effic WCAcc moorejt 2617 1 16 58 80 2 0 34 221 0 54 1896 6 11 25 1 02 1 04 0 14 99 52 100 00 zhong 1767 3 24 220 72 20 3 42 2306 5 65 1511 3 85 916 A S a 0 71 0 96 0 49 99 37 67 48 lui 2467 0 0 0 00 16 2 74 1970 4 82 1505 1 84 03 1 02 62335 O529 98296 5772 evans 3092 0 0 0 00 62 10 60 4960 12 14 1464 3 8 69 Dis 0 0 62 1 64 5 04 87 64 30 62 wengel 2430 2 64 824 90 1 Ode 767 1 88 630 3 33 4 E 0 18 0 18 4 26 99 63 0 40 mukho 2961 2 16 71 06 6 1 03 776 1 90 563 5 334m 04 31 0 82 0 20 93 15 30 228 jimenez 1449 1 16 302729 2 0 34 768 1 88 458 3 2672 55452 0 80 0 98 22315 GRO TOnZ neff 3194 0 0 0 00 74 12 65 669 1 64 352 5 2 09 10 0 0 50 1 93 0 51 96 03 32 60 cholik 1303 0 0 0 00 2 0 34 552 1 435 228739 16 gt s Le 3 07 259 35 99 69 66 270 jshoemak 2508 1 24 STZ 22 1 0 217 576 TiAl 229 01 1 3 6 SS35 0 55 0 55 3 74 99 20 39 20 kudo 2324 1 8 163 35 6 1 03 1152 2 82 211 1 Lei Ze t SSs 0 12 0 34 Da 965 TT 5 67 xztang 1835 1 8 18 99 176 3 1 05 10 0 SSa OO 02 ssa SS feller 1880 0 0 0 00 17 2391 170 0 42 148 1 088 095 1 83 0 14 97 59 84 31 maxi
41. Major Maui Structures Under Construction Nodes mnode_t Jobs myjob_t Reservations mres_t Partitions mpart_t QOS mqos_t Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ Appendix E Security E 1 Security Configuration Maui provides access control mechanisms to limit how the scheduling environment is managed The primary means of accomplishing this is through limiting the users and hosts which are trusted and have access to privileged commands and data With regards to users Maui breaks access into three distinct levels E 1 1 Level 1 Maui Admin Administrator Access A level 1 Maui Admin has global access to information and unlimited control over scheduling operations He is allowed to control scheduler configuration policies jobs reservations and all scheduling functions He is also granted access to all available statistics and state information Level 1 admins are specified using the ADMINI1 parameter E 1 2 Level 2 Maui Admin Operator Access Level 2 Maui Admins are specified using the ADMIN2 parameter The users listed under this parameter are allowed to change all job attributes and are granted access to all informational Maui commands E 1 3 Level 3 Maui Admin Help Desk Access Level 3 administrators users a specified via the ADMIN3 parameter They are allowed access to all informational Maui commands They cannot change scheduler or job attributes E 1
42. Maui RMTYPE 0 PBS should contact the associated RMHOST 0 cws RMPORT X lt INTEGER gt 0 resource manager The value 0 RMPORT 0 20001 specifies to use the appropriate default port for the resource Maui will attempt to contact the PBS server daemon manager type selected on host ews port 20001 specifies the host on which Maui should contact the RMTYPE 0 LL2 associated resource manager RMHOST 0 An empty value specifies to use RMPORT 0 0 RMSERVER X lt HOSTNAME gt NONE the default hostname for the resource manager selected NOTE this parameter is renamed RMHOST in Maui 3 0 6 and higher Maui will attempt to contact the Loadleveler version 2 Negotiator daemon on the default host and port as specified in the LL config files RMTIMEOUT 1 30 seconds maui will wait for a ae i RMTIMEOUT X lt INTEGER gt 15 response from the associated Maui will wait 30 seconds to receive a response from resource manager resource manager 1 before timing out and giving up Maui will try again on the next iteration RMTYPE O PBS RMHOST 0O clusterl ve T 0 15003 specifies type of resource RMPORT S manage Ovecomaces by RUEYEE Pas RMTYPEIX PBS or WIKI and RMSUBTYPE
43. QoS jobs will also receive priority reservations Use of this feature allows sites to maintain high throughput for important jobs by guaranteeing the a significant proportion of these jobs are making progress towards starting through use of the priority reservation A final reservation policy is in place to handle a number of real world issues Occasionally when a reservation becomes active and a job attempts to start various resource manager race conditions or corrupt state situations will prevent the job from starting By default Maui assumes the resource manager is corrupt releases the reservation and attempts to re create the reservation after a short timeout However in the interval between the reservation release and the re creation timeout other priority reservations may allocate the newly available resources reserving them before the original reservation gets an opportunity to reallocate them Thus when the original job reservation is re established its original resource may be unavailable and the resulting new reservation may be delayed several hours from the earlier start time The parameter RESERVATIONRETYTIME allows a site that is experiencing frequent resource manager race conditions and or corruption situations to tell Maui to hold on to the reserved resource for a period of time in an attempt to allow the resource manager to correct its state See also Reservation Overview Backfill Copyright 2000 2002 Supercluster Researc
44. Value Job and Resource Prioritization 5 1 Job Priority 5 2 Node Allocation Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 5 1 Job Prioritization In general prioritization is the process of determining which of many options best fulfills overall goals In the case of scheduling a site will often have multiple independent goals which may include maximizing system utilization giving preference to users in specific projects or making certain that no job sits in the queue for more than a given period of time The approach used by Maui in representing a multi facetted set of site goals is to assign weights to the various objectives so an overall value or priority can be associated with each potential scheduling decision With the jobs prioritized the scheduler can roughly fulfill site objectives by starting the jobs in priority order 5 1 1 Priority Overview 5 1 2 Priority Components 5 1 3 Common Priority Usage 5 1 4 Prioritization Strategies 5 1 5 Manual Priority Management Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 5 1 1 Priority Overview Maui s prioritization mechanism allows component and subcomponent weights to be associated with many aspects of a job so as to enable fine grained control over this aspect of scheduling To allow this level of control Maui uses a simple priority weighting hierarchy where the contribution of ea
45. When resources are available in more than one resource set the NODESETPRIORITYTYPE parameter allows control over how the best resource set is selected Legal values for this parameter are described in the table below Priority Type Description Details BESTFIT select the smallest minimizes fragmentation of larger resource resource set possible sets only supported when NODESETATTRIBUTE is set to PROCSPEED Selects the fastest possible nodes for the job select the resource set BESTRESOURCE with the best nodes select the resource set which will result in the minimal wasted Only supported when NODESETATTRIBUTE is set to PROCSPEED and NODESETTOLERANCE is gt 0 This algorithm is highly useful in environments with mixed speed compute nodes and a non load balancing parallel workload resources assuming no MINLOSS internal job load balancing is available assumes parallel jobs only run as fast as the slowest allocated node select the largest minimizes the creation of small resource set WORSTFIT resource set possible fragments but fragments larger resource sets On a per job basis each user can specify the equivalent of all parameters except NODESETDELAY As mentioned previously this is accomplished using the resource manager extensions See also N A Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 8 4 Preemption Policies enabled in Maui 3 0 7
46. a reservation could reserve 20 processors and 10 GB of memory for users Bob and John from Friday 6 00 AM to Saturday 10 00 PM Maui uses advance reservations extensively to manage backfill guarantee resource availability for active jobs allow service guarantees support deadlines and enable metascheduling Maui also supports both regularly recurring reservations and the creation of dynamic one time reservations for special needs Advance reservations are described in detail in the advance reservation overview 3 2 1 4 Policies Policies are generally specified via a config file and serve to control how and when jobs start Policies include job prioritization fairness policies fairshare configuration policies and scheduling policies 3 2 1 5 Resources Jobs nodes and reservations all deal with the abstract concept of a resource A resource in the Maui world is one of the following processors Processors are specified with a simple count value memory Real memory or RAM is specified in megabytes MB swap Virtual memory or swap is specified in megabytes MB disk Local disk is specified in megabytes MB In addition to these elementary resource types there are two higher level resource concepts used within Maui These are the task and the processor equivalent or PE 3 2 1 6 Task A task is a collection of elementary resources which must be allocated together within a single node For example a task may consist of o
47. about jobs Example gt diagnose r Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved ___ Accounts Fairshare Groups Job Frames Diagnose Priority Job Queue QOS Users profiler XXX INFO NOT YET AVAILABLE Purpose XXX Permissions This command can be run by any Maui Scheduler Administrator Parameters Flags Description Example Related Commands Default File Location u loadl bqs bin Notes None Copyright 1998 Maui High Performance Computing Center All rights reserved Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved releasehold releasehold hl al b JOBEXP Purpose Release hold on specified job s Permissions This command can be run by any Maui Scheduler Administrator Parameters JOBEXP Job expression of job s to release Flags a Release all types of holds user system batch for specified job s b Release batch hold from specified job s h Help for this command Description This command allows you to release batch holds or all holds system user and batch on specified jobs Any number of jobs may be released with this command Example 1 gt releasehold b frl7n02 1072 0 Batch hold released on all specified jobs In this example a batch hold was released from this one job Example 2 gt releasehold a frl7n02 1072 0 fr15n03 1017 0 All holds released on all specified jo
48. absolute or relative basis Only jobs with credentials listed in the reservation ACL i e USERLIST GROUPLIST can utilize the reserved resources However these jobs still have the freedom to utilize resources outside of the reservation The reservation will be assigned a name derived from the ACL specified If no reservation ACL is specified the reservation is created as a system reservation and no jobs will be allowed access to the resources during the specified timeframe valuable for system maintenance etc See the Reservation Overview for more information Reservations can be viewed using the showres command and can be released using the releaseres command Example 1 Reserve two nodes for use by users john and mary for a period of 8 hours starting in 24 hours setres u john mary s 24 00 00 d 8 00 00 TASKS 2 reservation john 1 created on 2 nodes 2 tasks node001 1 node005 1 Example 2 Schedule a system wide reservation to allow a system maintenance on Jun 20 8 00 AM until Jun 22 5 00 PM setres s 8 00 00_06 20 e 17 00 00_06 22 ALL reservation system 1 created on 8 nodes 8 tasks node001 node002 node003 node004 node005 node006 node007 node008 Hle e e FF FP He Example 3 Reserve one processor and 512 MB of memory on nodes node003 through node 006 for members of the group staff and jobs in the interactive class setres r PROCS 1 MEM 512 g staff l interactive node00 3 6
49. as the usage metric DEBITSUCCESSFULPE charge only for jobs which successfully complete using PE weighted wallclock time dedicated as the usage metric NOTE On systems where job wallclock limits are specified jobs which exceed their wallclock limits and are subsequently cancelled by the scheduler or resource manager will be considered as having successfully completed as far as charging is concerned even though the resource manager may report these jobs as having been removed or cancelled See also BANKTIMEOUT and BANKDEFERJOBONFAILURE Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved L 7 0 Controlling Resource Access Reservations Partitions and QoS Facilities 7 1 Advance Reservations 7 2 Partitions 73 QoS Facilities Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 7 1 Advance Reservations Reservation Overview An advance reservation is the mechanism by which Maui guarantees the availability of a set of resources at a particular time Every reservation consists of 3 major components a list of resources a timeframe and an access control list It is the job of the scheduler to make certain that the access control list is not violated during the reservation s lifetime i e its timeframe on the resources listed For example a reservation may specify that node002 is reserved for user Tom on Friday The scheduler will thus be cons
50. command Default File Location u loadl maui bin cancel job Notes None Copyright 1998 Maui High Performance Computing Center All rights reserved Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ changeparam Overview The changeparam command is used to dynamically change the value of any parameter which can be specified in the maui cfg file The changes take affect at the beginning of the next scheduling iteration They are not persistent only lasting until Maui is shutdown Format changeparam lt PARAMETER gt lt VALUE gt lt PARAMETER gt is any valid Maui parameter lt V ALUE gt is any valid value for lt PARAMETER gt Flags NONE Access This command can be run by any user with ADMINI authority Example Set Maui s LOGLEVEL to 6 for the current run gt changeparam LOGLEVEL 6 parameters changed Example Set Maui s ADMIN1 userlist to sys mike peter gt changeparam ADMINI sys mike peter parameters changed Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ checkjob checkjob ARGS lt OBID gt Purpose Display detailed job state information and diagnostic output for specified job Permissions This command can be run by any Maui admininstrator Additionally valid users may use this command to obtain information about their own jobs Args Details A provide output in the form of parsabl
51. eliminates the contention issue infinite length standing reservations cannot be made periodic One final example It was claimed earlier that access lists within a reservation are OR d together to determine reservation access However this rule has one notable exception triggered by use of the parameter SRMAXTIME This parameter controls the length of time a job can use the resources in a standing reservation This access mechanism can be AND d or OR d to the cumulative set of all other access lists as specified by the SRTIMELOGIC parameter Consider the following example configuration SRNAME 0 shortpool SRTASKCOUNT 0 32 SRPERIOD 0 WEEK SRWSTARTTIME 0 1 08 00 00 SRWENDTIME 0 5417400300 SRFEATURES 0 largememory SRMAXT IME 0 1 00 00 SRTIMELOGIC 0 AND SROOSLIST 0O high low special SRACCOUNTLIST 0O projectX projectY The term final example probably made it sound like we were just about finished didn t it So why are there 1 2 3 7 new parameters in this example Our apologies we figure only the die hards are reading at this point In a nutshell this specification asks for 32 tasks which translate to 32 nodes SRPERIOD states that this reservation is periodic on a weekly basis while the parameters SRWSTARTTIME and SRWENDTIME specify the week offsets when this reservation is to start and end In this case the reservation starts on Monday at 8 00 AM and runs until Friday at 5 00 PM The reservation is e
52. enabling what is called conservative backfill where every job which cannot run is given a reservation Most sites prefer the liberal backfill approach associated with the default RESERVATIONDEPTH of 1 or select a slightly higher value It is important to note that to prevent starvation in conjunction with reservations monotonically increasing priority factors such as queuetime or job xfactor should be enabled See the Prioritization Overview for more information on priority factors Another important consequence of backfill and reservation depth is its affect on job priority In Maui all jobs are prioritized Backfill allows jobs to be run out of order and thus to some extent job priority to be ignored This effect known as priority dilution can cause many site policies implemented via Maui prioritization policies to be ineffective Setting the RESERVATIONDEPTH parameter to a higher value will give job priority more teeth at the cost of slightly lower system utilization This lower utilization results from the constraints of these additional reservations decreasing the scheduler s freedom and its ability to find additional optimizing schedules Anecdotal evidence indicates that these utilization losses are fairly minor rarely exceeding 8 In addition to RESERVATIONDEPTH sites also have the ability to control how reservations are maintained Maui s dynamic job prioritization allows sites to prioritize jobs so that their priority orde
53. has a lot going on There will be several aspects of configuration however they are not too difficult individually First the queue structure The best place to handle this is via the PBS configuration Fire up qmgr and set up the nine queues described above PBS supports the node and walltime constraints as well as the queue priorities Maui will pick up and honor queue priorities configured within PBS Alternatively you can also specify these priorities directly within the Maui fs cfg file for resource managers which do not support this capability We will be using QBank to handle all allocations and so will want to configure the the per class charge rates there Note QBank 2 9 or higher is required for per class charge rate support Now two reservations are needed The first reservation will be for the 16 small memory nodes It should only allow node access to jobs requesting up to 16 processors In this environment this is probably most easily accomplished with a reservation class ACL containing the queues which allow 1 16 node jobs Monitoring Conclusions Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved Appendix B Extension Interface Maui supports an extension interface which allows external libraries to be linked to the Maui server This interface provides these libraries with full access to and control over all Maui objects and data It also allows this library the abilit
54. helpdesk or any member of the groups operations or sysadmin can use these resources While access is granted to the logical OR of access lists specified within a standing reservation access is only granted to the logical AND of access lists across different standing reservations Come again Compare standing reservations interactive and debug in the examples above Note that they both can allocate nodes node003 and node004 Assume that node003 had both of these reservations in place simultaneously and a job attempted to access this node during business hours when standing reservation interactive was active The job could only use the doubly reserved resources if it requested the run class interactiveAND it met the constraints of reservation debug i e was submitted by user he 1pdesk or by a member of the group operations or sysadmin 7 1 5 2 3 Reservation Stacking To make things just a little more confusing Maui will not stack reservations unless it has to If adequate resources exist it can allocate reserved resources side by side in a single SMP node rather than on top of each other Take the case of a 16 processor SMP node with two 8 processor standing reservations Eight of the processors on this node will be allocated to the first reservation and eight to the next Any configuration is possible The 16 processor nodes can also have 4 processors reserved for user John 10 processors reserved for group Staff with
55. idea of a task Maui uses the task concept extensively for its job and reservation management A task is simply an atomic collection of resources such as processors memory or local disk which must be found on the same node For example if a task requires 4 processors and 2 GB of memory Maui must find all processors AND memory on the same node it cannot allocate 3 processors and 1 GB on one node and 1 processor and 1 GB of memory on another node to satisfy this task Tasks constrain how Maui must collect resources for use in a standing reservation however they do not constrain the way in which Maui makes these cumulative resources available to jobs A job can use the resources covered by an accessible reservation in whatever way it needs If reservation X allocated 6 tasks with 2 processors and 512 MB of memory each it could support job Y which requires 10 tasks of 1 processor and 128 MB of memory or job Z which requires 2 tasks of 4 processors and 1 GM of memory each The task constraints used to acquire a reservation s resources are completely transparent to a job requesting use of these resources Task Count Using the task description the task count attribute defines how many tasks must be collected to satisfy the reservation request To create a reservation a task count and or a hostlist must be specified Hostlist A hostlist constrains the set of resource which are available to a reservation If no task count is specified the
56. in memory statistics which are loaded at scheduler start time from the scheduler checkpoint file See the Checkpoint Overview for more information This checkpoint file is updated from time to time and when the scheduler is shutdown allowing statistics to be collected over an extended timeframe At any time real time statistics can be reset using the resetstats command In addition to the showstats command the showgrid command also obtains its information from the in memory stats and checkpoint file This command display a processor time based matrix of scheduling performance for a wide variety of metrics Information such as backfill effectiveness or average job queue time can be determined on a job size duration basis See the showgrid command documentation for more information 9 2 2 Profiling Historical Usage Historical usage information can be obtained for a specific timeframe class of jobs and or portion of resources using the profiler command This command operates on the detailed job trace information recorded at the completion of each job These traces are stored in the directory pointed to by the STATDIR parameter which defaults to MAUIHOMEDIR stats Within this directory statistics files are maintained using the format WWW_MMM_DD_YYYY i e Mon_Jul_16_2001 with jobs traces being recorded in the file associated with the day the job completed Each job trace is white space delimited flat text and may be viewed directly with any
57. information from the resource manager s e Maui is able to properly start a new job Each of these areas are covered in greater detail below 2 3 1 Minimal Configuration Required To Start Up 2 3 1 1 Simulation Mode 23 1 2 Test Mode 23 1 3 Normal Mode 2 3 1 Minimal Configuration Required To Start Up Maui must have a number of parameters specified in order to properly start up There are three main approaches to setting up Maui on a new system These include the following 2 3 1 1 Simulation Mode Simulation mode is of value if you would simply like to test drive the scheduler or when you have a stable production system and you wish to evaluate how or even if the scheduler can improve your current scheduling environment An initial test drive simulation can be obtained via the following step gt vi maui cfg change SERVERMODE NORMAL to SERVERMODE SIMULATION add SIMRESOURCETRACEFILE traces Resource Trace1 add SIMWORKLOADTRACEFILE traces Workload Trace1 gt maul amp NOTE In simulation mode the scheduler does not background itself like it does in both TEST and NORMAL mode The sample workload and resource traces files allow the simulation to emulate a 192 node IBM SP In this mode all Maui commands can be run as if ona normal system The schedctl command can be used to advance the simulation through time The Simulation chapter describes the use of the simulator in detail If you are familiar
58. job queue specifies the credential component weight See Cred Factor NOTE thi t DIRECTSPECWEIGHT lt INTEGER gt 0 a NO 18 parameter DIRECTSPECWEIGHT 2 has been renamed CREDWEIGHT in Maui 3 0 7 and higher RESWEIGHT 10 specifies the priority weight to DISKWEIGHT 100 DISKWEIGHT lt INTEGER gt 0 De copes 1 Me amount oi dedicated disk space required per task by a job in MB if a job requires 12 tasks and 512 MB per task of dedicated local disk space Maui will increase the job s priority by 10 100 12 512 one or more of the following values space specifies flags which control DISPLAYFLAGS delimited NONE how maui client commands will DISPLAYFLAGS NODECENTRIC NODECENTRIC display various information DOWNNODEDELAYTIME 1 00 00 default time an unavailable node se DOWNNODEDELAYTIME DD HH MM SS 24 00 00 Down or Drain is marked Maui will assume down nodes will be available 1 unavailable hour after they go down unless a system reservation is placed on the node specifies whether or not Maui ENFORCERESOURCELIMITS one of the following ON or OFF OFF nonio take action when 4 job ELIMITS ON exceeds its requested resource usage FEATURENODETYPEHEADER xnt specifies the header used to ae i sp
59. load node specific information optional load job information load queue information optional cancel jobs which violate policies start jobs in accordance with available resources and policy constraints handle user commands oOo Nu Aa BW WV repeat Each step would complete before the next step started As systems continued to grow in size and complexity however it became apparent that the serial model described above would not work Three primary motivations drove the effort to replace the serial model with a concurrent threaded approach These motivations were reliability concurrency and responsiveness Reliability A number of the resource managers Maui interfaces to were unreliable to some extent This resulted in calls to resource management APT s with exitted or crashed taking the entire scheduler with them Use of a threaded approach would cause only the calling thread to fail allowing the master scheduling thread to recover Additionally a number of resource manager calls would hang indefinately locking up the scheduler These hangs could likewise be detected by the master scheduling thread and handled appropriately in a threaded environment Concurrency As resource managers grew in size the duration of each API global query call grew proportionally Particularly queries which required contact with each node individually became excessive as systems grew into the thousands of nodes A threaded interface al
60. lt STRING gt NONE NONE NOTE if no applicable value is specified the exact string NONE should be entered relative machine speed value a Sample Resource Trace COMPUTENODE AVAILABLE 0 cluster008 PBS1 423132 256 7140 2 1 1 1 LINUX62 Athlonk7 s950 compute batch 2 ethernet atm 1 67 NONE NONE NONE Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 16 3 Workload Traces Workload traces fully describe all scheduling relevant aspects of batch jobs including resources requested and utilized time of all major scheduling event i e submission time start time etc the job credentials used and the job execution environment Each job trace is composed of a single line consisting of 44 whitespace delimited fields as shown in the table below 4 16 3 1 Workload Trace Format gt 1632 Creating New Workload Traces 16 3 1 Workload Trace Format Field Name mie Data Format Default Value Details Index JobID 1 lt STRING gt No Name of job must be unique DEFAULT i Nodes 2 lt INTEGER gt 0 Number of nodes requested 0 no node request Requested count specified dasks 3 lt INTEGER gt 1 Number of tasks requested Requested User Name 4 lt STRING gt EN Name of user submitting job DEFAULT NO Group Name 5 lt STRING gt DEFAULT Primary group of user submitting job Wallclock Limit lt INTEGER gt 1 Maximum allowed job duration in seconds
61. more of the following fCORE fSCHED fSOCK fUI fLL fSDR fCONFIG fSTAT fSIM fSTRUCT fFS fCKPT fBANK fRM fPBS fWIKI fALL LOGFACILITY 00 10 00 log fALL specifies the length of time after which Maui will sync up a job s expected state with an unexpected reported state IMPORTANT NOTE Maui will not allow a job to run as long as its expected state does not match the state reported by the resource manager NOTE this parameter is named JOBS YNCDEADLINE in Maui 3 0 5 and earlier specifies the directory in which log files will be maintained If specified as a relative path LOGDIR will be relative to MAUIHOMEDIR see Logging Overview specifies which types of events to log see Logging Overview JOBSYNCT IME 00 01 00 LOGDIR tmp Maui will record its log files directly into the tmp directory LOGFACILITY RM PBS Maui will log only events involving general resource manager or PBS interface activities LOGFILE lt STRING gt LOGFILEMAXSIZE lt INTEGER gt maui log 10000000 name of the maui log file This file is maintained in the directory pointed to by lt LOGDIR gt unless lt LOGFILE gt is an absolute path see Logging Overview LOGFILE maui test log Log information will be written to the file maui test 1log located in the directory pointed to by the LOGDIR parameter maximum allowed size in bytes the log file before it will
62. one minute job being queued on average one minute while a 24 hour job would be queued an average of 24 hours Like queue time the effective XFactor subcomponent weight is the sum of two weights the XFWEIGHT parameter and the QOS specific XFWEIGHT setting For example the line QOSCFG special XFWEIGHT 5000 will cause jobs utilizing the QOS special to have their expansion factor subcomponent weight increased by 5000 5 1 2 4 3 Bypass BYPASS Subcomponent The bypass factor is the forgotten stepchild of the priority subcomponent family It was originally introduced to prevent backfill based starvation It is based on the bypass count of a job where the bypass count is increased by one every time the job is bypassed by a lower priority job via backfill The calculation for this factor is simply Over the years the anticipated backfill starvation has never been reported The good news is that if it ever shows up Maui is ready 5 1 2 5 Target Service TARG Component The target factor component of priority takes into account job scheduling performance targets Currently this is limited to target expansion factor and target queue time Unlike the expansion factor and queue time factors described earlier which increase gradually over time the target factor component is designed to grow exponentially as the target metric is approached This behavior causes the scheduler to do essentially all in its power to make certain the schedulin
63. reservation will attempt to reserve one task on each of the listed resources If a task count is specified which requests fewer resources than listed in the hostlist Maui will reserve only lt TASKCOUNTS tasks from the hostlist nodes If a taskcount is specified which requests more resources than listed in the hostlist Maui will reserve the hostlist nodes first and then seek additional resources outside of this list 7 1 5 1 4 Flags Reservation flags allow specification of special reservation attributes or behaviors The following flags are supported Flag Name Description reservation will only allow access to jobs which BYNAME meet reservation ACL s and explicitly request the resources of this reservation using the job ADVRES flag PREEMPTEE N A BESTEFFORT N A 7 1 5 2 Configuring Standing Reservations Standing reservations allow resources to be dedicated for particular uses at a regular time of day and or time of week There is extensive applicability of standing reservations for everything from daily dedicated job runs to improved use of resources on weekends All standing reservation attributes are specified via parameters An overview of standing reservation capabilities is included below followed by a series of examples 7 1 5 2 1 Standing Reservation Overview A standing reservation is similar to a normal administrative reservation in that it also places an access control list on a specified set of resources Res
64. session key which f Maui will use to communicate CLIENTKEY silverB 0x3325584 CLIENTKEY lt X gt lt INTEGER gt NONE n am peer daemon l Maui will use the session key 0x3325584 for this parameter may only lencrypting and decrypting messages communicated be specified in the from silverB maui private cfg config file time which Maui client commands will wait for a CLIENTTIMEOUT 00 15 00 CLIENTTIMEOUT DD HH MM SS 00 00 30 response from the Maui server Maui clients will wait up to 15 minutes for a response NOTE may also be specified from the server before timing out as an environment variable specifies the credential component weight See Cred CREDWEIGHT lt INTEGER gt 1 Factor NOTE this parameter OREDWEIGHT 2 was named DIRECTSPECWEIGHT prior to Maui 3 0 7 specifies the default classes space delimited list of one or more supported on each node for RM z DEFAULTCLASSLIST lt STRING gt s NONE systems which do not provide DEFAULTCLASSLIST serial parallel this information specifies the number of times a DEFERCOUNT lt INTEGER gt 24 job can be deferred before it will JDEFERCOUNT 12 be placed in batch hold specifies number of time a job DEFERSTARTCOUNT lt INTEGER gt 1 will be allowed to fail in its start DEFERSTARTCOUNT 3 attempts before being deferred specifies amount of time a job DEFERTIME DD HH MM SS 1 00 00 will be held in the deferred state orpRSTARTTIME 0 05 00 before being released back to the Idle
65. sites to specify various policies by adjusting the class initiator to node mapping For example a site running serial jobs may want to allow a particular 8 processor node to run any combination of batch and special jobs subject to the following constraints only 8 jobs of any type allowed simultaneously no more than 4 special jobs allowed simultaneously To enable this policy the site may set the node s MAXJOB policy to 8 and configure the node with 4 special class initiators and 8 batch class initiators Note that in virtually all cases jobs have a one to one correspondence between processors requested and class initiators required However this is not a requirement and with special configuration sites may choose to associate job tasks with arbitrary combinations of class initiator requirements In displaying class initiator status Maui signifies the type and number of class initiators available using the format lt CLASSNAME gt lt CLASSCOUNT gt This is most commonly seen in the output of node status commands indicating the number of configured and available class initiators or in job status commands when displaying class initiator requirements Arbitrary Resource Node can also be configured to support various arbitrary resources Information about such resources can be specified using the NODECFG parameter For example a node may be configured to have 256 MB RAM 4 processors 1 GB Swap and 2 tape drives Co
66. so partition boundaries are often valuable if a underlying network topology make certain resource allocations undesirable Additionally per partition policies can be specified to grant control over how scheduling is handled on a partition by partition basis See the Partition Overview for more information 12 1 2 Frames Frame based location information is orthogonal to the partition based configuration and is mainly an organizational construct In general frame based location usage a node is assigned both a frame and a slot number This approach has descended from the IBM SP2 organizational approach in which a frame can contain any number of slots but typically contains between 1 and 64 Using the frame and slot number combo individual compute nodes can be grouped and displayed in a more ordered manner in certain Maui commands i e showstate Currently frame information can only be specified directly by the system via the SDR interface on SP2 Loadleveler systems In all other systems this information must be manually specified via the NODECFG parameter Example maui cfg NODECFG node024 FRAME 1 SLOT 1 NODECFG node025 FRAME 1 SLOT 2 NODECFG node026 FRAME 2 SLOT 1 PARTITION special When specifying node and frame information slot values must be in the range of 1 to 32 limited to 1 to 16 in Maui 3 0 and earlier and frames must be in the range of 1 to 64 12 1 3 Queues Some resource managers allow queues or classes to be d
67. space delimited list of user names NONE NONE users listed under the parameter ADMIN2 are allowed to change all job attributes and are granted access to all informational Maui commands Valid values include user names or the keyword ALL users listed under the parameter ADMIN3 are allowed access to all informational maui commands They cannot change scheduler or job attributes Valid values include user names or the keyword ALL ADMIN2 jack karen jack and karen can modify jobs i e canceljob setqos setspri etc and can run all Maui information commands ADMIN3 ops user ops can run all informational command such as checkjob or checknode specifies the number idle jobs to evaluate for backfill The backfill algorithm will evaluate BACKFILLDEPTH 128 BACKFILLDEPTH lt INTEGER gt 0 no limit the top lt X gt priority jobs for evaluate only the top 128 highest priority idle jobs for scheduling By default all jobs consideration for backfill are evaluated specifies the criteria used by the one of the following PROCS backfill algorithm to determine BACKFILLMETRIC PROCSECONDS SECONDS PE or PROCS the best jobs to backfill Only BACKFILLMETRIC PROCSECONDS PESECONDS applicable when using BESTFIT o
68. the job actually ran before completing and releasing its allocated resources For example a job with a wallclock limit of 1 hour will be given the need resources for up to an hour but may complete in only 20 minutes The output of showg r shows when the job will complete if it runs up to its specified wallclock limit In the simulation jobs actually complete when their recorded runtime is reached Let s look at this job more closely gt checkjob fr8n01 804 0 We can wee that this job has a wallclock limit of 5 minutes and requires 5 nodes We can also see exactly which nodes have been allocated to this job There is a lot of additional information which the checkjob man page describes in more detail Let s advance the simulation another step gt schedctl S Look at the queue again to see if anything has happened gt showg r No surprises Everything is one minute closer to completion gt schedctl S gt showg r Job fr8n01 804 0 is still 2 minutes away from completing as expected but notice that both jobs fr8n01 191 0 and fr8n01 189 0 have completed early Although they had almost 24 hours remaining of wallclock limit they terminated In reality they probably failed on the real world system where the trace file was being created Their completion freed up 40 processors which the scheduler was able to immediately use by starting two more jobs Let s lo
69. the job is S tarting R unning or Ddle Start Relative start time of the reservation Time is displayed in HH MM SS notation and is relative to the present time End Relative end time of the reservation Time is displayed in HH MM SS notation and is relative to the present time Reservation that will not complete in 1 000 hours are marked with the keyword INFINITY Duration Duration of the reservation in HH MM SS notation Reservations lasting more than 1 000 hours are marked with the keyword INFINITY Nodes Number of nodes involved in reservation StartTime Time Reservation became active Example 2 gt showres n Reservations on Sat Dec 14 08 31 09 NodeName Type ReservationID JobState Start Duration StartTime frl0nl1 mhpcc edu Job fr4n02 126 0 Starting 0 02 00 6 00 00 Sat Dec 14 08 29 09 fr26n01 mhpcc edu Job fr4n02 126 0 Starting 0 02 00 6 00 00 Sat Dec 14 08 29 09 Fr5n09 mhpcc edu Job fr4n02 126 0 Starting 0 02 00 6 00 00 Sat Dec 14 08 29 09 System SYSTEM 0 N A 20 00 00 10 00 00 Sun Dec 15 04 31 09 frl8n15 mhpcc edu Job fr4n02 126 0 Starting 0 02 00 6 00 00 Sat Dec 14 08 29 09 fr20n02 mhpcc edu Job fr4n02 126 0 Starting 0 02 00 6 00 00 Sat Dec 14 08 29 09 User load1 0 N A 0 00 00 30 00 00 Sat Dec 14 08 31 09 Group daf 0 N A 10 00 00 INFINITE Sat Dec 14 18 31 09 fr20n15 mhpcc edu Job fr4n02 126 0 Starting 0 02 00 6 0
70. the less likely the system is to be overcommitted however a larger value also means more resources are in this reserve which are unavailable for scheduling The right solution is to migrate the users over to the batch system or provide them with a constrained resource box to play in either through a processor partition another system or via a logical software system The value in the box is that it prevents this unpredictable background load from wreaking havoc with an otherwise sane dedicated resource reservation system Maui can reserve resource for jobs according to all info currently available However the unpredictable nature of the background load may mean those resources are not available when they should be resulting in cancelled reservations and the inability to enforce site policies and priorities The second aspect of this environment which must be monitored is the trade off between high job throughput and job starvation The locally greedy approach of favoring the smallest shortest jobs will have a negative effect on larger and longer jobs The large long jobs which have been queued for some time can be pushed to the front of the queue by increasing the QUEUETIMEWEIGHT factor until a satisfactory balance is achieved Conclusions Mixed batch non batch systems are very very nasty Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ A 4 Case Study Standard Production SP Und
71. the minimum duration of jobs to be displayed in matrix PLOTMINTIME 1 00 00 PLOTMAXTIME 64 00 00 peer pe pee ia aed outputs as displayed by the each matrix output will display data in columns for showerid or profiler commands jobs requesting between 1 and 64 hours of run time specifies the number of rows PLOTMINPROC 1 into which the range of PLOTMAXPROC 1024 processors requested per job will PLOTPROCSCALE 1 0 PLOTPROCSCALE lt INTEGER gt 9 be divided when disp layed in each matrix output will display job data divided into matrix outputs as displayed by 10 rows which are evenly spaced geometrically the showgrid or profiler covering the range of jobs requesting between and commands 1024 processors PLOTMINTIME 2 00 00 specifies the number of columns PLOTMAXTIME 32 00 00 into which the range of job PLOTTIMESCALE 5 PLOTTIMESCALE lt INTEGER gt 11 ae b each matrix output will display job data divided into 5 La olaged be the shononid or columns which are evenly spaced geometrically 1sp i yed by the showgric covering the range of jobs requesting between 2 and 32 profiler commands hours i e display columns for 2 4 8 16 and 32 hours of walltime eo of the following specifies how preemptible jobs PREEMPTIONPOLICY CHECKPOINT PREEMPTIONPOLICY REQUEUE SUSPEND CHECKPOINT REQUEUE will be preempted Available in jobs that are to be preempted will be checkpointed and
72. to the currently running Maui and is not propagated to the config file Changes can also be made by modifying the config file and restarting the scheduler or issuing schedctl R which forces the scheduler to basically recycle itself Let s stop at an even number iteration 60 gt schedctl s 601 The s flag indicates that the scheduler should stop at the specified iteration gt showstats v This command may hang a while as the scheduler simulates up to iteration 60 The output of this command shows us the 21 jobs have now completed Currently only 191 of the 195 nodes are busy Lets find out why the 4 nodes are idle First look at the idle jobs gt showg i The output shows us that there are a number of single processor jobs which require between 10 hours and over a day of time Lets look at one of these jobs more closely gt checkjob fr1n04 2008 0 If a job is not running checkjob will try to determine why it isn t At the bottom of the command output you will see a line labeled Rejection Reasons It states that of the 195 nodes in the system the job could not run on 191 of them because they were in the wrong state i e busy running other jobs and 4 nodes could not be used because the configured memory on the node did not meet the jobs requirements Looking at the checkjob output further we see that this job requested nodes with gt 512 MB of
73. used to favor jobs associated with a particular project while QUEUETIME can be used to favor those jobs which have been waiting the longest Queue Time Expansion Factor Resource Fairshare Cred Target Metrics Each priority factor group may contain one or more subfactors For example the Resource factor consists of Node Processor Memory Swap Disk and PE components Figure lt X gt shows the current priority breakdown From the figure it is quickly apparent that the prioritization problem is fairly nasty due to the fact that every site needs to prioritize a bit differently Fortunately there has not yet been a site that has desired to use more than a fraction of these priority factors thus greatly simplifying the job priority tuning issue When calculating a priority the various priority factors are summed and then bounded between 0 and MAX_PRIO_VAL which is currently defined as 100000000 one billion Each priority factor is reviewed in detail below The command diagnose p is designed to assist in visualizing the priority distribution resulting from the current job priority configuration Also the showgrid command will help indicate the impact of the current priority settings on scheduler service distributions Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 5 1 5 Manual Job Priority Adjustment Batch administrator s regularly find a need to adjust the calculate
74. with Maui you may wish to use the simulator to tune scheduling policies for your own workload and system The profiler tool can be used to obtain both resource and workload traces and is described further in the section Collecting Traces Generally at least a week s worth of workload should be collected to make the results of the simulation statistically meaningful Once the traces are collected the simulation can be started with some initial policy settings Typically the scheduler is able to simulate between 10 and 100 minutes of wallclock time per second for medium to large systems As the simulation proceeds various statistics can be monitored if desired At any point the simulation can be ended and the statistics of interest recorded One or more policies can be modified the simulation re run and the results compared Once you are satisfied with the scheduling results the scheduler can be run live with the tuned policies 2 3 1 2 Test Mode Test mode allows you to evaluate new versions of the scheduler on the side In test mode the scheduler connects to the resource manager s and obtains live resource and workload information Using the policies specified in the maui cfg file the test mode Maui behaves identical to a live normal mode Maui except the code to start cancel and pre empt jobs is disabled This allows you to exercise all scheduler code paths and diagnose the scheduling state using the various diagnostic client co
75. 0 00 Sat Dec 14 08 29 09 User load1l 0 N A 0 00 00 30 00 00 Sat Dec 14 08 31 09 Group daf 0 N A 10 00 00 INFINITE Sat Dec 14 18 31 09 fr26nl1l mhpcc edu Job Fr4n02 126 0 Starting 0 02 00 6 00 00 Sat Dec 14 08 29 09 frl7nll mhpcc edu Job Fr4n02 126 0 Starting 0 02 00 6 00 00 Sat Dec 14 08 29 09 fr25n12 mhpcc edu Job Fr4n02 126 0 Starting 0 02 00 6 00 00 Sat Dec 14 08 29 09 fr26n1l6 mhpcc edu Job Fr4n02 126 0 Starting 0 02 00 6 00 00 Sat Dec 14 08 29 09 fr5nl12 mhpcc edu Job Fr4n02 126 0 Starting 0 02 00 6 00 00 Sat Dec 14 08 29 09 System SYSTEM 0 N A 20 00 00 10 00 00 Sun Dec 15 04 31 09 Fr5nl15 mhpcc edu Job fr4n02 126 0 Starting 0 02 00 6 00 00 Sat Dec 14 08 29 09 This example shows reservations for nodes The fields are as follows NodeName Node on which reservation is placed Reservation Type This will be one of the following Job User Group Account or System This is the name of the reservation Job reservation names are identical to the job name User Group or Account reservations are the user group or account name followed by a number System This field is valid only for job reservations It indicates the state of the job associated with the Relative start time of the reservation Time is displayed in HH MM SS notation and is relative to Duration of the reservation in HH MM SS notation Reservations lasting more than 1000 hours are Duration Nodes StartTime Type ReservationI
76. 0 17 04 00 34 08 00 TOTAL a 16 1 01 2 64 128 256 T TOT 1 01 2 1 00 3 1 24 2 1 12 2 ll Job Weighted X Factor 1 0888 Node Weighted X Factor 1 1147 NS Weighted X Factor 1 1900 Total Samples 9 The showgrid command returns a table with data for the specified STASTICTYPE parameter The left most column shows the maximum number of nodes required by the jobs shown in the other columns The column heads indicate the maximum wall clock time in HH MM SS notation requested by the jobs shown in the columns The data returned in the table varies by the STATISTICTYPE requested For table entries with one number it is of the data requested For table entries with two numbers the left number is the data requested and the right number is the number of jobs used to calculate the average Table entries that contain only dashes indicate no job has completed that matches the profile associated for this inquiry The bottom row shows the totals for each column Following each table is a summary which varies by the STATISTICTYPE requested This particular example shows the average expansion factor grid Each table entry indicates two pieces of information the average expansion factor for all jobs that meet this slot s profile and the number of jobs that were used to calculate this average For exam
77. 1 hour Job C is the next highest priority job and requires 1 processor for 2 hours Maui examines the jobs and correctly determines that job A must finish in 2 hours and thus the earliest job B can start is in 2 hours Maui also determines that job C can start and finish in less than this amount of time Consequently Maui starts job C on the idle processor One hour later job A completes early Apparently the user overestimated the amount of time his job would need by a few hours Since job B is now the highest priority job it should be able to run However job C a lower priority job was started an hour ago and the resources needed for job B are not available Maui re evaluates job B s reservation and determines that it can be slid forward an hour At time 3 job B starts Ok now the post game show Job A is happy because it ran to completion Job C is happy because it got to start immediately Job B is sort of happy because it got to run 1 hour sooner than it originally was told it could However if backfill was not enabled job B would have been able to run 2 hours earlier Not a big deal usually However the scenario described above actually occurs fairly frequently This is because the user estimates for how long their jobs will take is generally very bad Job wallclock estimate accuracy or wallclock accuracy is defined as the ratio of wall time required to actually run the job divided by the wall time requested for the job Wallclock accu
78. 16 CLASSCFG batch MAXNODE 32 QOSCFG hiprio MAXJOB 3 OMAXNODE 64 This configuration is identical to the line above with the exception of the final QOSCFG line This line does two things Only 3 highprio jobs may run simultaneously highprio QOS jobs may run with up to 64 nodes per credential ignoring other credential MAXNODE limits Given the above configuration assume a job was now submitted with the credentials user steve group staff class batch and QOS hiprio This job will be allowed to start so long as running it does not lead to any of the following conditions total nodes used by user steve jobs do not exceed 64 total active jobs associated with user steve does not exceed 2 total active jobs associated with group staff does not exceed 5 total nodes dedicated to class batch jobs do not exceed 64 total active jobs associated with QOS hiprio does not exceed 3 While the above example is a bit complicated for actual use at most sites similar combinations may be needed to enforce site policies on many larger systems Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 6 3 FairShare 6 3 1 Overview 6 3 2 FairShare Parameters 6 3 3 Using Fairshare Information 6 3 1 Overview Fairshare is a mechanism which allows historical resource utilization information to be incorporated into job feasibility and priority decisions Maui s fairshare implementation
79. 2 Obtain information about the simulated compute resources in the file traces Resource Tracel 3 Obtain information about the jobs to be run in simulation in the file traces Workload Tracel1 4 Load the job and node info start whatever jobs can be started on the nodes and then wait for user commands Do not advance simulated time until instructed to do so gt maui amp give the scheduler a few seconds to warm up and then look at the list of jobs currently in the queue To obtain a full description of each of the commands used below please see the command s man page gt showq This command breaks the jobs in the queue into three groups Active jobs which are currently running Idle jobs which can run as soon as the required resources become available and Non Queued jobs which are currently ineligible to be run because they violate some configured policy By default the simulator initially submits 100 jobs from the workload trace file Workload Tracel Looking at the showq output it can be seen that the simulator was able to start 11 of these jobs on the 195 nodes described in the resource trace file Resource Tracel Look at the running jobs more closely gt showg r The output is sorted by job completion time We can see that the first job will complete in 5 minutes Look at the initial statistics to see how well the scheduler is doing gt
80. 2000 2002 Supercluster Research and Development Group All Rights Reserved resetstats resetstats h Purpose Resets statistics to start up state Permissions This command can be run by any Maui Scheduler Administrator Parameters None Flags h Help for this command Description This command resets all internally stored Maui Scheduler statistics to the initial start up state as of the time the command was executed Example o 3 resetstats Statistics Reset at time Wed Feb 25 23 24 55 1998 Related Commands None Default File Location u loadl maui bin resetstats Notes None Copyright 1998 Maui High Performance Computing Center All rights reserved runjob runjob ARGS lt JOBID gt Purpose Immediately run the specified job Permissions This command can be run by any Maui administrator Parameters JOBID Name of the job to run Args C f h n lt NODELIST gt p lt PARTITION gt S Description Description Clear job parameters from previous runs used to clear PBS neednodes attribute after PBS job launch failure Attempt to force the job to run ignoring throttling policies Help for this command Attempt to start the job using the specified nodelist where nodenames are comma or colon delimited Attempt to start the job in the specified partition Attempt to suspend the job Attempt to force the job to run ignoring throttling policies QoS constaint
81. 29 11 36 16 fr28n01 1851 0 vertex Running 1 19 09 49 Fri Aug 29 11 42 16 fr1l7n11 1380 0 vertex Running 1 19 41 22 Fri Aug 29 12 13 49 fr17n16 1533 0 vertex Running 1 20 04 32 Fri Aug 29 12 36 59 fr17n06 1502 0 vertex Running 1 20 16 24 Fri Aug 29 12 48 51 fr17n10 1466 0 wengel Running 32 20 24 04 Fri Aug 29 10 58 11 fr28n13 701 0 kudo Running 8 20 25 46 Fri Aug 29 10 58 13 f r28n03 1689 0 vertex Running 1 20 50 31 Fri Aug 29 13 22 58 fr28n13 631 0 vertex Running I 21 17 40 Fri Aug 29 13 50 07 fr28n13 708 0 yshi Running 8 222 A9210 Eri Aug 29 13321237 fr17n05 1395 0 yshi Running 8 23 36 36 Fri Aug 29 14 09 03 frl7n11 1388 0 jshoemak Running 24 23 51 10 Fri Aug 29 14 23 37 fr28n07 2304 0 rich00l Running 1 26 09 44 Fri Aug 29 13 42 11 fr28n11 3091 0 rampi Running 1 26297200 9 Fri Aug 29 053292257 36 Active Jobs 251 of 254 Processors Active Efficiency 98 82 IDLE JOBS JOBNAME USERNAME STATE PROC CPULIMIT QUEUETIME fr28n03 1718 0 ozturan Idle 64 0 16 40 Thu Aug 28 22 25 48 fr17n03 1430 0 jason Idle 128 2 00 00 Wed Aug 27 00 56 49 fr17n08 1331 0 jason Idle 128 2 00 00 Wed Aug 27 00 56 21 frl17n15 1393 0 moraiti Idle 128 3 20 00 Fri Aug 29 09 58 56 fr17n09 534 0 kdeacon Idle 64 1 00 00 Fri Aug 29 04 38 48 fr28n13 697 0 jpark Idle 16 24 00 00 Fri Aug 29 03 44 45 fr17n07 1499 0 jpark Idle 16 24 00 00 Fri Aug 29 04 42
82. 3 2 and later SRGROUPLIST 1 staff ops special specifies the groups which will SRCLASSLIST 1 interactive SRGROUPLIST X one or more space delimited group names ALL be allowed access to this Maui will allow jobs with the listed group ID s or standing reservation which request the job class interactive to use the resources covered by standing reservation 1 specifies the set of host from which Maui can search for SRHOSTLIST 3 node001 node002 node003 resources to satisfy the SRRESOURCES 3 PROCS 2 MEM 512 reservation If SRTASKCOUNT 3 2 SRHOSTLIST X one or more space delimited host names ALL SRTASKCOUNT is also SRMAXTIME X DD HH MM SS SRNAME X lt STRING gt 1 no time based access NONE specified only lt SRTASKCOUNTS tasks will be reserved Otherwise all hosts listed will be reserved specifies the maximum allowed overlap between a the standing reservation and a job requesting resource access specifies name of standing reservation lt X gt Maui will reserve 2 tasks with 2 processors and 512 MB each using resources located on node001 node002 and or node003 SRMAXTIME 6 1 00 00 Maui will allow jobs to access up to one hour of resources in standing reservation 6 SRNAME 1 interactive The name of standing reservation 1 is interactive specifies the partition in which SRPARTITION 0 OLD SRPARTITION
83. 4 Admininstrative Hosts If specified the ADMINHOST parameter allows a site to specify a subset of trusted hosts All administrative commands level 1 3 will be rejected unless they are received from one of the hosts listed E 2 Interface Security As part of the U S Department of Energy SSS Initiative Maui interface security is being enhanced to allow full encryption of data and GSI like security If these mechanisms are not enabled Maui also provides a simple secret checksum based security model Under this model each client request is packaged with the client ID a timestamp and a checksum of the entire request generated using a secret site selected key checksum seed This Key is selected when the Maui configure script is run and may be regenerated at any time by rerunning configure and rebuilding Maui E 2 1 Interface Development Notes Details about the checksum generation algorithm can be found in the Socket Protocol Description document Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved L Appendix F Maui Parameters See the Parameters Overview in the Maui Admin Manual for further information about specifying parameters Name ACCOUNTCEFG lt ACCOUNTID gt Format list of zero or more space delimited lt ATTR gt lt V ALUE gt pairs where lt ATTR gt is one of the following PRIORITY FSTARGET QLIST QDEF PLIST PDEF FLAGS or a fairness policy specific
84. 6 The key at the bottom of the system map can be used to determine unusual node states For example fr7n15 is currently in the state down After the key a series of warning messages may be displayed indicating possible system problems In this case warning message indicate that there are memory problems on three nodes fr3n07 fr4n06 and fr4n09 Also warning messages indicate that job fr15n09 1097 0 is having difficulty starting Node fr11n08 is in state BUSY but has no job assigned to it It possibly has a runaway job running on it Related Commands None Default File Location u loadl maui bin showstate Notes None Copyright 1998 Maui High Performance Computing Center All rights reserved Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved showstats showstats FLAGS Purpose Show resource usage statistics Access This command can be run by any Maui level 1 2 or 3 Administrator Parameters NONE Flags NOTE this command supports all generic maui command flags Flag Description O O Z a lt ACCOUNTID gt display account statistics g lt GROUPID gt display group statistics OO RISNODED gt isplay node statisties L SE display summary information NOTE only valid with the n flag gt ns ae a e u lt USERID gt display user statistics v display verbose information Description This command shows various acc
85. A 64 proc O2K system needs to be scheduled with a significant background load Resources Compute Nodes 64 processor 32 GB O2K system Resource Manager OpenPBS 2 3 Network InternalSGI network Workload Job Size range in size from 1 to 32 processors Job Length jobs range in length from 15 minutes to 48 hours Job Owners various NOTES This is a login development machine meaning at any given time there may be a significant load originating with jobs processes outside of the resource manager s view or control The major scheduling relevant impact of this is in the area of cpu load and real memory consumption Constraints Must do The scheduler must run the machine at maximum capacity without overcommitting either memory or processors A significant and variable background load exists from jobs submitted outside of the resource manager s view or control The scheduler must track and account for this load and allow space for some variability and growth of this load over time The scheduler should also kill any job which violates its requested resource allocation and notify the associated user of this violation Goals Should do The scheduler should maximize the throughput associated with the queued jobs while avoiding starvation as a secondary concern Analysis The background load causes many problems in any mixed batch interactive environment One problem which will occur results from the fact that a situation may aris
86. API On a regular basis use the command grep E WARNINGIALERTIERROR maui log to get a listing of the problems the scheduler is detecting On a production system working normally this list should usually turn up empty The messages are usually self explanatory but if not viewing the log can give context to the message If a problem is occurring early when starting the Maui Scheduler before the configuration file is read maui can be started up using the L LOGLEVEL flag If this is the first flag on the command line then the LOGLEVEL is set to the specified level immediately before any setup processing is done and additional logging will be recorded If problems are detected in the use of one of the client commands the client command can be re issued with the L lt LOGLEVEL gt command line arg specified This argument causes debug information to be logged to STDERR as the client command is running Again lt LOGLEVEL gt values from 0 to 9 are supported In addition to the log file the Maui Scheduler reports all events it determines to be critical to the UNIX syslog facility via the daemon facility using priorities ranging from INFO to ERROR This logging is not affected by LOGLEVEL In addition to errors and critical events all user commands that affect the state of the jobs nodes or the scheduler are also logged via syslog The logging information is extremely helpful in diagnosing problems but it can also be useful if
87. Access Determining who can use which partition is specified using the CFG parameters USERCFG GROUPCFG ACCOUNTCFG QOSCFG CLASSCFG and SYSTEMCFG These parameters allow both a partition access list and default partition to be selected on a credential or system wide basis using the PLIST and PDEF keywords By default the access associated with any given job is the logical or of all partition access lists assigned to the job s credentials Assume a site with two partitions general and test The site management would like everybody to use the general partition by default However one user steve needs to perform the majority of his work on the test partition Two special groups staff and mgmt will also need access to use the test partition from time to time but will perform most of their work in the general partition The example configuration below will enable the needed user and group access and defaults for this site USERCFG DEFAULT PLIST general USERCFG steve PLIST general test PDEF test GROUPCFG staff PLIST general test PDEF general GROUPCFG mgmt PLIST general test PDEF general Note that the DEFAULT user has no default partition specified If only a single partition is provided in the access list it will be selected as the default partition In Maui 3 0 6 and earlier partition access would be controlled using the following stanza in the fairshare config file fs cfg USER DEFAULT PLIST general USER steve PL
88. CTION specifies the functions to be ARAEFUNCIION SS TRING NONE trapped UpdateNodeUtilization GetNodeSResTim TRAPJOB lt STRING gt NONE specifies the jobs to be trapped TRAPJOB buffy 0023 0 TRAPNODE lt STRING gt NONE specifies the nodes to be trapped TRAPNODE node001 node004 node005 TRAPRES lt STRING gt NONE Lace the reservations tobe J RAPRES interactive G1 specifies the weight assigned to USAGEWEIGHT lt INTEGER gt 0 the percent and total job usage USAGEWEIGHT 100 subfactors USAGEPERCENTWEIGHT lt INTEGER gt specifies whether or not job USEMACHINESPEED ON USEMACHINESPEED ON or OFF OFF Larue i eri be dof job lt X gt specifying a wallclock limit of 1 00 00 scaled by the machine speed of yWould be given only 40 minutes to run if started on a the node s they are running on node with a machine speed of 1 5 list of zero or more space delimited specifies user specific l lt ATTR gt lt V ALUE gt pairs where lt ATTR gt attributes See the flag USERCFG john MAXJOB 50 QDEF highprio is one of the following overview for a description of USERCFG lt USERID gt NONE ee up to 50 jobs submitted under the user ID john will PRIORITY FSTARGET QLIST QDEF legal flag values be allowed to execute simultaneously and will be PLIST PDEF FLAGS or a fairness policy NOTE Only av
89. D reservations are given the name SYSTEM followed by a number JobState reservation Start the present time Duration marked with the keyword INFINITY StartTime Time Reservation became active Example 3 gt showres fr35n08 3360 0 Reservations Type ReservationID S Start End Job fr35n08 3360 0 S 8 24 06 15 35 54 1 reservation located In this example information for a specific reservation job is displayed See Also setres create new reservations releaseres release existing reservations diagnose r diagnose view the state of existing reservations Reservation Overview description of reservations and their use Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 24 00 00 16 Thu Mar 5 03 08 38 showstart showstart h lt JOBID gt Purpose Display the earliest possible start and completion times for a specified job Permissions This command can be run by any user Parameters JOBID Job to be checked Flags h Help for this command Description This command displays the earliest possible start time of a job If the job already possesses a reservation the start time of this reservation will be reported If no such reservation exists this command will determine the earliest time a reservation would be created assuming this job was highest priority If this job does not have a reservation and it is not highest priority the value of returned information may b
90. Ded FSTet AvgXF MaxXF AvgQH Effic WCAcc Account Number Number of running jobs Number of processors allocated to running jobs Number of proc hours required to complete running jobs Number of jobs completed Percentage of total jobs that were completed by account Total proc hours requested by completed jobs Percentage of total proc hours requested by completed jobs that were requested by account Total proc hours dedicated to active and completed jobs The proc hours dedicated to a job are calculated by multiplying the number of allocated procs by the length of time the procs were allocated regardless of the job s CPU usage Percentage of total proc hours dedicated that were dedicated by account Fairshare target An account s fairshare target is specified in the fs cfg file This value should be compared to the account s node hour dedicated percentage to determine if the target is being met Average expansion factor for jobs completed A job s XFactor expansion factor is calculated by the following formula QueuedTime RunTime WallClockLimit Highest expansion factor received by jobs completed Average queue time in hours of jobs Average job efficiency Job efficiency is calculated by dividing the actual node hours of CPU time used by the job by the node hours allocated to the job Average wall clock accuracy for jobs completed Wall clock accuracy is calculated by dividing a job s actual ru
91. EDICATED ADVRES HOSTLIST RESTARTABLE specified job flags on all jobs o i SIMDEFAULTJOBELAGS PREEMPTEE DEDICATED NONE supplied in the workload trace Maui will set the DEDICATED job flag on all jobs PREEMPTOR file loaded from the workload trace file iteration on which a Maui SIMEXITITERATION lt INTEGER gt 0 no exit iteration simulation will create a SIMEXITITERATION 36000 simulation summary and exit Beroon More Or the Lolowne controls how Maui handles trace E eee SIMFLAGS IGNHOSTLIST IGNCLASS IGNQOS NONE based information Maui will ignore hostlist information specified in the IGNMODE IGNFEATURES workload trace file SIMIGNOREJOBFLAGS 4 gt NONE job flags if supplied in the willi oF PREEMPTEE DEDICATED Maui will ignore the DEDICATED job flag if PREEMPTOR workload trace file specified in any job trace specifies how many jobs the SIMINITIALQUEUEDEPTH 64 SIMJOBSUBMISSIONPOLICY CONSTANTJOBDEPTH SIMINITIALQUEUEDEPTH lt INTEGER gt 16 simulator will initially place in Maui will initially place 64 idle jobs in the queue and the idle job queue because of the specified queue policy will attempt to maintain this many jobs in the idle queue throughout the duration of the simulation specifies how the simulator will submit new jobs into the idle queue NORMAL mode causes jobs to be submitted at the time one of the followin
92. EIGHT lt INTEGER gt 0 nodecount is specified by the NODEWEIGHT 1000 user job If the job only specifies tasks or processors no node factor will be applied to the job s total priority This will be rectified in future versions specifies the name of the m NOTIFICATIONPROGRAM lt STRING gt NONE recat handloall Seer nce eLiinge tal Sd atk tools notifyme pl notification call outs specifies the coefficient to be Senin 0 10 ido PEWEIGHT X lt INTEGER gt 0 multiplied by job s PE a processor equivalent priority each job s priority will be increased by 10 100 its factor PE factor specifies the maximum number IOTMINPROC 1 of processors requested by jobs p OTMAXPROC 1024 PLOTMAXPROC lt INTEGER gt 512 to be displayed in matrix outputs as displayed by the showgrid or each matrix output will display data in rows for jobs profiler commands requesting between 1 and 1024 processors specifies the maximum duration P LOTMINTIME 0 0 g PLOTMAXTIME DD HH MM SS 68 00 00 pf jobs to be displayed in matrix P OTe outp uts as displayed by the each matrix output will display data in columns for showgrid or profiler commands jobs requesting between 1 and 64 hours of run time specifies the minimum number p IOTMINPROC 1 of processors requested by jobs LOTMAXPROC 1024 1 to be displayed in matrix outputs PLOTMINPROC lt INTEGER gt profiler commands requesting between 1 and 1024 processors specifies
93. GROUPCFG staff MAXJOB 5 CLASSCFG DEFAULT MAXNODE 16 CLASSCFG batch MAXNODE 32 This configuration may potentially apply multiple limits to a single job User steve limits will cause that jobs submitted under his user ID will be constrained so that he may only run up to 2 simultaneous jobs with an aggregate node consumption of 30 nodes However if he submits a job to a class other than batch he may be limited further Only 16 total nodes may be used simultaneously by jobs running in any given class with the exception of the class batch If steve submitted a job to run in the class interactive for example and there were jobs already running in this class using a total of 14 nodes his job would be blocked unless it requested 2 or fewer nodes by the default limit of 16 nodes per class 6 2 1 2 Multi Dimension Fairness Policies Multi dimensional fairness policies allow a site to specify policies based on combinations of job credentials A common example might be setting a maximum number of jobs allowed per queue per user or a total number of processors per group per QoS As with basic fairness policies multi dimension policies are specified using the CFG parameters Early versions of Maui 3 2 enabled the following multi dimensional fairness policies MAXJOB Class User MAXNODE Class User MAXPROC Class User These limits would be specified in the following manner CLASSCFG X MAXJOBPERUSER lt LIMIT gt CLASSCFG X MAXNODEPERUSE
94. IOACCRUALPOLICY allows a site to select how a job will accrue queue time based on meeting various throttling policies Regardless of the policy used to determine a job s queue time this effective queue time is used in the calculation of the QUEUETIME XFACTOR TARGETQUEUETIME and TARGETXFACTOR priority subcomponent values The need for a distinct effective queue time is necessitated by the fact that most sites have pretty smart users and pretty smart users like to work the system whatever system it happens to be A common practice at some long existent sites is for some users to submit a large number of jobs and then place them on hold These jobs remain with a hold in place for an extended period of time and when the user is ready to run a job the needed executable and data files are linked into place and the hold released on one of these pre submitted jobs The extended hold time guarantees that this job is now the highest priority job and will be the next to run The use of the JOBPRIOACCRUALPOLICY parameter can prevent this practice as well as preventing queue stuffers from doing similar things on a shorter time scale These queue stuffer users submit hundreds of jobs at once so as to swamp the machine and hog use of the available compute resources This parameter prevents the user from gaining any advantage from stuffing the queue by not allowing these jobs to accumulate any queue time based priority until they meet certain id
95. IST general test PDEF test GROUP staff PLIST general test PDEF general GROUP mgmt PLIST general test PDEF general 7 2 3 Requesting Partitions Users may request to use any partition they have access to on a per job basis This is accomplished using the resource manager extensions since most native batch systems do not support the partition concept For example on a PBS system a job submitted by a member of the group staff could request that the job run in the test partition by adding the line PBS W x PARTITION test to the command file See the resource manager extension overview for more information on configuring and utilizing resource manager extensions 7 2 4 Miscellaneous Partition Issues Special jobs may be allowed to span the resources of multiple partitions if desired by associating the job with a QOS which has the flag SPAN set See the QOSCFG parameter A brief caution use of partitions has been quite limited in recent years as other more effective approaches are selected for site scheduling policies Consequently some aspects of partitions have received only minor testing Still note that partitions are fully supported and any problem found will be rectified See Also N A Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 7 3 Quality of Service QoS Facilities 73 1 QoS Overview 73 2 QoS Enabled Privileges 73 2 1 Special Prioritization 7 3 2 2 Service Acce
96. JOBWALLTIME _ DD HH MM SS 10 00 00 00 which do not explicitly set this value Maui will assign a wallclock limit of 1 day to jobs which do not explicitly specify a wallclock limit SYSTEMMAXPROCPERJOB lt INTEGER gt 1 NO LIMIT specifies the maximum number of processors that can be requested by any single job SYSTEMMAXJOBPROC 256 Maui will reject jobs requesting more than 256 processors SYSTEMMA XPROCSECONDPERJOB SYSTEMMAXJOBWALLTIME TARGWEIGHT lt INTEGER gt DD HH MM SS lt INTEGER gt 1 NO LIMIT 1 NO LIMIT specifies the maximum number of proc seconds that can be requested by any single job specifies the maximum amount of wallclock time that can be requested by any single job SYSTEMMAXJOBPROCSECOND 86400 Maui will reject jobs requesting more than 86400 procs seconds i e 64 processors 30 minutes will be rejected while a 2 processor 12 hour job will be allowed to run SYSTEMMAXJOBWALLTIME 1 00 00 00 Maui will reject jobs requesting more than one day of walltime specifies the weight to be applied to a job s queuetime and expansion factor target components TARGETWEIGHT 1000 TASKDISTRIBUTIONPOLICY one of DEFAULT or LOCAL DEFAULT specifies how job tasks should be mapped to allocated resources TASKDISTRIBUTIONPOLICY DEFAULT Maui should use standard task distribution algorithms TRAPFUN
97. MAXJOBPERUSER 1 NODECFG DEFAULT MAXLOAD 2 5 Also See lt N A gt Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 13 0 Resource Managers and Interfaces 13 1 Resource Manager Overview 13 2 Resource Manager Configuration i 13 3 Resource Manager Extensions 13 4 Adding Resource Manager Interfaces Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 13 1 Resource Manager Overview Maui requires the services of a resource manager in order to properly function This resource manager provides information about the state of compute resources nodes and workload jobs Maui also depends on the resource manager to manage jobs instructing it when to start and or cancel jobs Maui can be configured to manage one or more resource managers simultaneously even resource managers of different types However migration of jobs from one resource manager to another is not currently allowed meaning jobs submitted onto one resource manager cannot run on the resources of another 13 1 1 Scheduler Resource Manager Interactions 3 1 1 1 Resource Manager Commands 3 1 1 2 Resource Manager Flow 13 1 2 Resource Manager Specific Details Limitations Special Features 13 1 1 Scheduler Resource Manager Interactions Maui interacts with all resource managers in the same basic format Interfaces are created to translate Maui concepts regarding workload and resour
98. Maui Administrator s Guide Maui 3 2 Last Updated May 16 Overview The Maui Scheduler can be thought of as a policy engine which allows sites control over when where and how resources such as processors memory and disk are allocated to jobs In addition to this control it also provides mechanisms which help to intelligently optimize the use of these resources monitor system performance help diagnose problems and generally manage the system Table of Contents 1 0 Philosophy and Goals of the Maui Scheduler 2 0 Installation and Initial Configuration 2 1 Building and Installing Maui 2 2 Initial Configuration 2 3 Initial Testing 3 0 Maui Basics 3 1 Layout of Maui Components 3 2 Scheduling Environment and Objects 3 3 Scheduling Iterations and Job Flow 3 4 Configuring the Scheduler 4 0 Maui Commands 4 1 Client Overview 4 2 Monitoring System Status 4 3 Managing Jobs 4 4 Managing Reservations 4 5 Configuring Policies 4 6 End User Commands 4 7 Miscellaneous Commands 5 0 Assigning Value Job and Resource Prioritization 5 1 Job Priority 5 2 Node Allocation ag Policies Fairshare and Allocation Management 7 0 Controlling Resource Access Reservations Partitions and QoS Facilities 72 Par it ons emption sting and Simulation 1B J ee ATE Extensions 13 4 Adding Resource Manager Interfaces 14 0 Trouble Shooting and System Maintenance 14 1 Internal Diagnostics 14 2 Logging Facilities 14 3 Usin
99. MemoryRequestedByJob TotalConfiguredMemory DiskRequestedByJob TotalConfiguredDisk SwapRequestedByJob TotalConfiguredSwap TotalConfiguredProcs For example say a job requested 20 of the total processors and 50 of the total memory of a 128 processor MPP system Only two such jobs could be supported by this system The job is essentially using 50 of all available resources since the system can only be scheduled to its most constrained resource in this case memory The processor equivalents for this job should be 50 of the processors or PE 64 Let s make the calculation concrete with one further example Assume a homogeneous 100 node system with 4 processors and 1 GB of memory per node A job is submitted requesting 2 processors and 768 MB of memory The PE for this job would be calculated as PE MAX 2 100 4 768 100 1024 100 4 3 This result makes sense since the job would be consuming 3 4 of the memory on a 4 processor node The calculation works equally well on homogeneous or heterogeneous systems uniprocessor or large way SMP systems 3 2 1 8 Class or Queue A class or queue is a logical container object which can be used to implicitly or explicitly apply policies to jobs In most cases a class is defined and configured within the resource manager and associated with one or more of the following attributes or constraints Attribute Attribute Description essses sSS Dekult J
100. NONE PREEMPTEE PREEMPTOR NOQUEUE associates ee FLAGS ADVRES indicates a exact set superset or subset of HOSTLIST comma delimited list of hostnames NONE HOSTLIST nodeA nodeB nodeE resource allocation NODESET lt SETTYPE gt lt SETATTR gt lt SETLIST gt NONE NODESET ONEOF PROCSPEED 350 400 450 specifies the partition or PARTITION math geology PARTITION lt STRING gt lt STRING gt NONE have access The job must only run in the math partition or the ae geology partition based on system wide or credential based partition access lists a QOS STRINGS ROKE lo OS highprio Indicates QUEUEJOB FALSE one of FALSE or TRUE TRUE SGE lt WINDOWCOUNT gt lt DISPLAYNAME gt NONE SCE 8 pinky SID lt STRING gt NONE sS siD silverA 2 TPN lt INTEGER gt fo ren TRL SO lt INTEGER gt lt INTEGER gt OQ sf sR 2 4 8 16 If more than one extension is required in a given job extensions can be concatenated with a semicolon separator using the format lt ATTR gt lt VALUE gt lt ATTR gt lt VALUE gt See the following examples Example 1 Loadleveler command file comment HOSTLIST nodel node2 Q0S special SID silverA Job must run on nodes nodel and node2 using the QoS special The job is also associated with the system id silverA allowing the silver daemon to mo
101. R lt LIMIT gt CLASSCFG X MAXPROCPERUSER lt LIMIT gt Later versions of Maui will allow more generalized use of these limits using the following syntax lt O1 gt CFG lt OID1 gt MAX lt A1 gt lt O2 gt lt OID2 gt lt LIMIT gt Where O1 is one of the following objects USER GROUP ACCOUNT QOS or CLASS A1 is one of the following attributes JOB PROC PS PE WC NODE or MEM O2 is one of the following objects USER GROUP ACCOUNT QOS or CLASS If OID2 is specified the limit is applied only to that object instance Otherwise the limit is applied to all appropriate objects by default The following examples may clarify CLASSCFG batch MAXJOB 3 MAXNODE USER 8 Allow class batch to run up the 3 simultaneous jobs Allow any user to use up to 8 total nodes within class batch CLASSCFG fast MAXPROC USER steve 3 MAXPROC USER bob 4 Allow users steve and bob to use up to 3 and 4 total processors respectively within class fast See Also N A 6 2 2 Override Limits Like all job credentials the QOS object may be also be associated with resource usage limits However this credential can also be given special override limits which supersede the limits of other credentials Override limits are applied by preceding the limit specification with the letter O The configuration below provides an example of this USERCFG steve MAXJOB 2 MAXNODE 30 GROUPCFG staff MAXJOB 5 CLASSCFG DEFAULT MAXNODE
102. RESCTLPOLICY is set to ANY Description This command displays all reservations currently in place within the Maui Scheduler The default behavior is to display reservations on a reservation by reservation basis Example 1 gt showres Reservations Type ReservationID S Start End Duration Nodes StartTime Job fr4n01 902 0 5 0 02 00 0 08 00 0 10 00 16 Sat Dec 14 08 29 09 Job fr5nli1 776 0 S 0 01 00 1 59 00 2 00 00 8 Sat Dec 14 08 30 09 Job fr5nii1 177 0 E 0 01 00 0 02 20 0 03 20 1 Sat Dec 14 08 30 09 Job r5n12 179 0 3 0 00 30 15930 2 00 00 3 Sat Dec 14 08 30 39 Job fr5n12 180 0 0 00 30 02 29 30 0 30 00 4 Sat Dec 14 08 30 39 Job fr5n13 155 0 0 00 00 2 00 00 2 00 00 4 Sat Dec 14 08 31 09 Group daf 0 10 00 00 INFINITY INFINITY 16 Sat Dec 14 18 31 09 User loadl 0 0 00 00 30 00 00 30 00 00 16 Sat Dec 14 08 31 09 System SYSTEM 0 20 00 00 30 00 00 10 00 00 40 Sun Dec 15 04 31 09 25 Reservations Located This example shows all reservations on the system The fields are as follows Type Reservation Type This will be one of the following Job User Group Account or System ReservationID This is the name of the reservation Job reservation names are identical to the job name User Group or Account reservations are the user group or account name followed by a number System reservations are given the name SYSTEM followed by a number S State This field is valid only for job reservations It indicates whether
103. TEGER gt 0 no stop iteration stop and was for a command to Maui should stop after the first iteration of simulated resume scheduling scheduling and wait for admin commands determines wall time speedup SIMTIMERAT IO 10 SIMTIMERATIO lt INTEGER gt 0 no time ratio Simulated Maui time will Maui simulation time will advance 10 times faster advance lt SIMTIMERATIO gt faster than real wall time than real world wall time For example in 1 hour Maui will process 10 hours of simulated workload SIMWORKLOADTRACEFILE lt STRING gt SRACCESS X DEDICATED or SHARED SRACCOUNTLIST X list of valid account names SRCHARGEACCOUNT X any valid accountname one or more of the following lt ATTR gt lt VALUE gt pairs ACCOUNTLIST CHARGEACCOUNT CLASSLIST DAYS DEPTH ENDTIME FLAGS GROUPLIST HOSTLIST JOBATTRLIST NODEFEA TURES SRCFG X PARTITION PERIOD PRIORITY PROCLIMIT QOSLIST RESOURCES STARTTIME TASKCOUNT TASKLIMIT TIMELIMIT TPN USERLIST WENDTIME WSTARTTIME traces workload trace DEDICATED NONE NONE NONE specifies the file from which maui will obtain job information S ITMWORKLOADTRACEFILE traces jobs 2 when running in simulation mode Maui will attempt to Maui will obtain job traces when running in locate the file relative to simulation mode from the lt MAUIHOMEDIR gt unless lt MAUIHOMEDIR gt traces jobs 2 file specified as an absolute path
104. TH FSINTERVAL and FSDECAY can be freely and dynamically modified such changes may result in unexpected fairshare status for a period of time as the fairshare data files with the old FSINTERVAL setting are rolled out 6 3 3 Using Fairshare Information With the mechanism used to determine current fairshare usage explained above we can now move on that actually using this information As mentioned in the fairshare overview fairshare information primarily used in determining the fairshare priority factor This factor is actually calculated by determining the difference between the actual fairshare usage of an entity and a specified target usage See Also The diagnose f command was created to allow diagnosis and monitoring of the fairshare facility FairShare Prioritization vs Hard FairShare Enforcement FSENFORCEMENT Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 6 4 Allocation Management Overview An allocations manager also known as an allocations bank or cpu bank is a software system which manages resource allocations where a resource allocation grants a job a right to use a particular amount of resources This is not the right place for a full allocations manager overview but a brief review may point out the value in using such a system An allocations manager functions much like a bank in that it provides a form a currency which allows jobs to run on an HPC system The owners o
105. TIMEWEIGHT parameter is used in conjunction with the USERCFG line to specify a queue time target of 4 hours Assume now that the site decided that it liked this priority mix but they had a problem with users cheating by submitting large numbers very short jobs They would do this because very short jobs would tend to have rapidly growing xfactor values and would consequently quickly jump to the head of the queue In this case a factor cap would be appropriate These caps allow a site to say I would like this priority factor to contribute to a job s priority but only within a defined range This prevents certain priority factors from swamping others Caps can be applied to either priority components or subcomponents and are specified using the lt COMPONENTNAMESCAP parameter i e QUEUETIMECAP RESCAP SERVCAP etc Note that both component and subcomponent caps apply to the pre weighted value as in the following equation Priority CIWEIGHT MIN C1CAP SUM SLIWEIGHT MIN S11CAP S11S SLZWELGHT MIN SIL2ZCAP S125 C2WEIGHT MIN C2CAP SUM S21WEIGHT MIN S21CAP S21S S22WEIGHT MIN S22CAP S22S s Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 5 1 4 Prioritization Strategies Each component or subcomponent may be used to accomplish different objectives WALLTIME can be used to favor or disfavor jobs based on their duration Likewise ACCOUNT can be
106. TRING gt NONE access to the reserved resources name set to first ee name listed in ACL NAME lt STRING gt or SYSTEM if no oak Hon ACL specified partition in which resources must be located list of QOS s that will be allowed access to the PARTITION lt STRING gt ANY NONE reserved resource colon delimited list of zer or more of specifies the the following lt ATTR gt lt VALUE gt resources to pairs be reserved PROCS lt INTEGER gt PROCS 1 per task 1 MEM lt INTEGER gt indicates all DISK lt INTEGER gt resources on SWAP lt INTEGER gt specifies the QOS_LIST lt STRING gt lt STRING gt available should be expression is specified Maui will apply the reservation ALL or TASKS i gt lt TASKCOUNT gt or lt HOST_REGEX gt regardless of existing reservations and exclusitivity issues If TASKS is used Maui will only allocate accessible resources absolute or relative time reservation will start list of users that will be allowed access to the reserved resources HH MM SS _MO DD YY or NOW DD HH MM SS STARTTIME lt STRING gt lt STRING gt Description The setres command allows an arbitrary block of resources to be reserved for use by jobs which meet the specifed access constraints The timeframe covered by the reservation can be specified on either an
107. Time job started running StartTime 0 0 jo 0 3 3 3 3 i 3 7 2 8 5 9 23 2 9 1 1 0 0 3 4 4 3 5 227 EZT 2s 343 2353 oTa Oe a PA 2 18 24 1 230 7963 02 OD 226 236231 SADA 581 5ga d rea Or O98 23 3 242 293 36 oe 56 35 45 51 08 33 5 58 16 37 39 6 6 al 3 07 03 37 11 27 After displaying the running jobs a summary is provided indicating the number of jobs the number of allocated processors and the system utilization Example 3 o showq i SystemQueueTime fr28n03 1718 22 25 48 fri7n03 1430 18 29 26 fr28n13 634 18 30 04 fr28n09 32 18 32 58 fri7n15 1393 09 58 56 frl7n09 534 04 38 48 f r28n13 697 03 44 45 frl17n07 1499 04 42 31 frl7n06 1517 06 45 46 fr28n13 706 1045353 fril7n16 1550 10 53 54 fril7n12 1528 12211230 fr28n09 50 14 01 59 Fr28n09 51 14 07 16 frl7n16 1551 14 22 09 Jobs 15 The fields are as follows JobName JobName Priority XFactor Q User Group Nodes WCLimit Class SystemQueueTime Time job was admitted into the system queue An asterisk at the end of a job job fr28n03 1718 0 in this example indicates that the job has a job reservation created for it The details of this reservation can be displayed using the check job command 0 0 Total BackLog Priority XFactor 97615272 59 0 125372 60 125365 11 0 118071 7 0 110712 2 64 68841 10 9
108. X lt STRING gt ALL the standing reservation should gnty select resource for standing reservation 0 in be created partition OLD m EE ER SRPERIOD 1 WEEK SRPERIOD X one of DAY WEEK or INFINITY DAY jae the periodicity of the standing reservation each standing reservation covers a one week period specifies that jobs with the listed SRQOSLIST A Sa zero or more valid QOS names NONE QOS names can access the SRRESOURCES X SRSTARTTIME X semicolon delimited lt ATTR gt lt VALUE gt pairs HH MM SS PROCS 1 All processors available on node 00 00 00 maui will allow jobs using QOS 1 3 4 and 5 to use reserved resources the reserved resources specifies what resources constitute a single standing reservation task each task must be able to obtain all of its resources as an atomic unit on a single node Supported resources currently include the following SRRESOURCES 1 PROCS 1 MEM 512 each standing reservation task will reserve one processor and 512 MB of real memory PROCS number of processors MEM real memory in MB DISK local disk in MB SWAP virtual memory in MB SRSTARTTIME 1 08 00 00 SRENDTIME 1 17 00 00 specifies the time of day the standing reservation becomes active standing reservation 1 is active from 8 00 AM until 5 00 PM SRTASKCOUNT X SRTIMELOGIC X lt INTEGER g
109. XXX b XXX b XXX Frame 28 G XXX XXX D XXX XXX D XXX Frame 29 A C A A C A C XXxXxxx Key XXX Unknown Down w Job Down Key a Any lower cas Check Memory on Node fr3n07 Check Memory on Node fr4n06 Check Memory on Node fr4n09 Active Job 1 fr15n09 1097 0 Starting Active Job 1 fr15n09 1097 0 Starting Active Job 1 fr15n09 1097 0 Starting Active Job 1 fr15n09 1097 0 Starting Active Job 1 fr15n09 1097 0 Starting Node friin08 is Busy but Has No Job Scheduled XXX XXX XXX XXXXX b D X Has Has Has Has Has b b D XX Idle w Job Node Node Node Node Node XXX XXX XXX XXXX XXX XXX XXX XXXX b b X z idle Busy w No Job letter indicates an idle node that is assigned to a job Drained fr7n09 Allocated which is in state fri5n01 Allocated which is fril5n03 Allocated which is fri5n05 Allocated which is fri5n07 Allocated which is in in in in state state state state Idle rdle Tdle Idle Idile In this example nine active jobs are running on the system Each job listed in the top of the output is associated with a letter For example job fr17n11 942 0 is associated with the letter A This letter can now be used to determine where the job is currently running By looking at the system map it can be found that job fr17n11 942 0 Gob A is running on nodes fr2n10 fr2n13 fr2n16 fr3n0
110. __ Maui stats files in format stats lt Y Y Y Y gt _ lt MM gt _ lt DD gt l___ Maui fairshare data files in format FS lt EPOCHTIME gt __ tools directory for local tools called by Maui OPTIONAL BY DEFAULT __ traces directory for Maui simulation trace files REQUIRED FOR SIMULATIONS ___ resource trace1 sample resource trace file _____ workload trace1 sample workload trace file l__ bin directory for Maui executable files REQUIRED BY DEFAULT maui Maui scheduler executable _ maui_client Maui user interface client executable ___ profiler tool used to analyze Maui statistics _ssre directory for Maui source code files REQUIRED FOR BUILD __ spool directory for temporary Maui files REQUIRED FOR ADVANCED FEATURES ___ contrib directory containing contributed code in the areas of GUI s algorithms policies etc MAUIINSTDIR __ bin directory for installed Maui executables maui Maui scheduler executable maui_client Maui user interface client executable profiler tool used to analyze Maui statistics etc maui cfg optional file This file is used to override default MAUIHOMEDIR settings it should contain the string MAUIHOMEDIR DIRECTORY to override the built in MAUIHOMEDIR setting When Maui is configured via the configure script the user is queried for the location of the Maui home directory and this directory MAUIHOMEDIR is compiled in as the defaul
111. a 2936 0 0 0 00 1 0 17 191 0 47 129 1 OlT 15 0 88 0 88 4 49 99 84 69 10 ktgnov71 2838 0 0 0 00 1 0 17 192 0 47 95 55 oS 0 53 0 53 0 34 90 07 51 20 This example shows a statistical listing of all active users The top line User Statistics Initialized of the output indicates the timeframe covered by the displayed statistics The statistical output is divided into two statistics categories Running and Completed Running statistics include information about jobs that are currently running Completed statistics are compiled using historical information from both running and completed jobs The fields are as follows UserName Name of user UID User ID of user Jobs Number of running jobs Procs Number of procs allocated to running jobs ProcHours Number of proc hours required to complete running jobs Jobs Number of jobs completed Percentage of total jobs that were completed by user PHReq Total proc hours requested by completed jobs o Percentage of total proc hours requested by completed jobs that were requested by user PHDed Total proc hours dedicated to active and completed jobs The proc hours dedicated to a job are calculated by multiplying the number of allocated procs by the length of time the procs were allocated regardless of the job s CPU usage Percentage of total prochours dedicated that were dedicated by user FSTgt Fairshare target A user s fairshare target is specified in the fs cfg file This value s
112. abilities Finally the RMNMPORT allows specification of the resource manager s node manager port and is only required when this port has been set to a non default value It is currently only used within PBS to allow MOM specific information to be gathered and utilized by Maui 13 1 2 Scheduler Resource Manager Interactions In the simplest configuration Maui interacts with the resource manager using the four primary functions listed below GETJOBINFO Collect detailed state and requirement information about idle running and recently completed jobs GETNODEINFO Collect detailed state information about idle busy and defined nodes STARTJOB Immediately start a specific job on a particular set of nodes CANCELJOB Immediately cancel a specific job regardless of job state Using these four simple commands Maui enables nearly its entire suite of scheduling functions More detailed information about resource manager specific requirements and semantics for each of these commands can be found in the specific resource manager overviews LL PBS or WIKI In addition to these base commands other commands are required to support advanced features such a dynamic job support suspend resume gang scheduling and scheduler initiated checkpoint restart More information about these commands will be forthcoming Information on creation a new scheduler resource manager interface can be found in the Adding New Resource Manager Interfaces sect
113. able resources by running jobs out of order When Maui schedules it prioritizes the jobs in the queue according to a number of factors and then orders the jobs into a highest priority first sorted list It starts the jobs one by one stepping through the priority list until it reaches a job which it cannot start Because all jobs and reservations possess a start time and a wallclock limit Maui can determine the completion time of all jobs in the queue Consequently Maui can also determine the earliest the needed resources will become available for the highest priority job to start Backfill operates based on this earliest job start information Because Maui knows the earliest the highest priority job can start and which resources it will need at that time it can also determine which jobs can be started without delaying this job Enabling backfill allows the scheduler to start other lower priority jobs so long as they do not delay the highest priority job If Backfill is enabled Maui protects the highest priority job s start time by creating a job reservation to reserve the needed resources at the appropriate time Maui then can any job which not not interfere with this reservation Backfill offers significant scheduler performance improvement In a typical large system enabling backfill will increase system utilization by around 20 and improve turnaround time by an even greater amount Because of the way it works essentially filling in
114. agnose j Display high level summary of job attributes and perform sanity check on job attributes state diagnose q Display various reasons job is considered blocked or non queued showbf v Determine general resource availability subject to specified constraints See also Diagnosing System Behavior Problems Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 15 0 Improving User Effectiveness 15 1 User Feedback Loops 15 2 User Level Statistics 15 3 Enhancing Wallclock Limit Estimates 15 4 Providing Resource Availability Information 15 5 Job Start Time Estimates Ve 15 6 Collecting Performance Information on Individual Jobs Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 15 1 User Feedback Loops In evaluating a given system it is interesting to note that almost invariably real world systems outperform simulated systems Even when all policies reservations workload and resource distributions are fully captured and emulated What is it about real world usage that is not emulated via a simulation The answer is the user feedback loop the impact of users making decisions to optimize their level of service based on real time information A user feedback loop is created any time information is provided to a user which modifies his job submission or job management behavior As in a market economy the cumulative effect of many us
115. ailable in Maui assigned the QOS highprio by default specification 3 0 7 and higher specifies the weight assigned to USERWEIGHT lt INTEGER gt 0 the specified user priority see USERWEIGHT 100 Credential Priority Factor specifies whether or not job prioritization should be based on the time the job has been eligible to run i e idle and USESYSTEMQUEUETIME OFF meets all fairness policies ON the queuetime and expansion factor components of a USES YSTEMQUEUETIME ON or OFF OFF or a aa a been idle ljob s priority will be calculated based on the length of OFF NOTE In Maui 3 0 8 time the job has been in the idle state and higher this parameter has See QUEUETIMEFACTOR for more info been superseded by the JOBPRIOACCRUALPOLICY parameter specifies the maximum total pre weighted contribution to job priority which can be XECAP 10000 XFCAP lt DOUBLE gt 0 NO CAP contributed by the expansion __ Maui will not allow a job s pre weighted XFactor factor component This value 18 priority component to exceed the value 10000 specified as an absolute priority value not as a percent XFMINWCLIMIT 0 01 00 specifies the minimum job wallclock limit that will be jobs requesting less than one minute of wallclock time XFMINWCLIMIT DD HH MM SS 1 NO LIMIT will be treated as if their wallclock limit was set to one considered in job expansion factor priority calculations minute when determining expansion factor for priority ca
116. all subroutines are logged at higher LOGLEVELs Example CheckPolicies fr4n01 923 0 2 Reason 2 Status Information Information about internal status is logged at all LOGLEVELs Critical internal status is indicated at low LOGLEVELs while less critical and voluminous status information is logged at higher LOGLEVELs Example INFO Job fr4n01 923 0 Rejected Max User Jobs INFO Job 25 fr4n01 923 0 Rejected MaxJobPerUser Policy Failure 3 Scheduler Warnings Warnings are logged when the scheduler detects an unexpected value or receives an unexpected result from a system call or subroutine They are not necessarily indicative of problems and are not catastrophic to the scheduler Example WARNING cannot open fairshare data file home load1 maui stats FS 87000 4 Scheduler Alerts Alerts are logged when the scheduler detects events of an unexpected nature which may indicate problems in other systems or in objects They are typically of a more severe nature than are warnings and possibly should be brought to the attention of scheduler administrators Example ALERT job fr5n02 202 0 cannot run deferring job for 360 Seconds 5 Schedulers Errors Errors are logged when the scheduler detects problems of a nature of which it is not prepared to deal It will try to back out and recover as best it can but will not necessarily succeed Errors should definitely be be monitored by administrators Example ERROR cannot connect to Loadleveler
117. allation consists of the following steps 2 1 Maui Installation 2 2 Initial Maui Configuration 2 3 Testing Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 2 1 Maui Installation Building Maui To install Maui untar the distribution file enter the maui lt VERSIONS gt directory then run configure and make as shown in the example below gt gtar xzvf maui 3 0 7 tar gz gt cd maui 3 0 7 gt configure gt make Installing Maui Optional When you are ready to use Maui in production you may install it into the install directory you have configured using make install gt make install Note Until the install step is performed all Maui executables will be placed in MAUIHOMEDIR bin i e maui 3 0 7 bin in the above example Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 2 2 Initial Maui Configuration After you install Maui there are a few decisions which must be made and some corresponding information which will need to be provided in the Maui configuration file maui cfg The configure script automatically sets most of these parameters for you However this document provides some additional information to allow further initial configuration If you are satisfied with the values specified in configure then you can probably skip this section The parameters needed for proper initial startup include the following
118. allows site administrators to set system utilization targets for users groups accounts classes and QOS levels Administrators can also specify the timeframe over which resource utilization is evaluated in determining whether or not the goal is being reached This timeframe is indicated by configuring parameters specifying the number and length of fairshare windows which are evaluated to determine historical resource usage Fairshare targets can then be specified for those credentials 1 e user group class etc which administrators wish to have affected by this information 6 3 2 Fairshare Parameters Fairshare is configured at two levels First at a system level configuration is required to determine how fairshare usage information is to be collected and processed Secondly some configuration is required on a per credential basis to determine how this fairshare information affects particular jobs The system level parameters are listed below FSINTERVAL specifies the timeframe covered by each fairshare window FSDEPTH specifies the number of windows to be evaluated when determining current fairshare utilization FSDECAY specifies the decay factor to be applied to fairshare windows FSPOLICY specifies the metric to use when tracking fairshare usage if set to NONE fairshare information will not be used for either job prioritization or job feasibility evaluation FSCONFIGFILE specifies the name of the file which contains the
119. ally all critical messages while a log level of 4 will provide general information describing all actions taken by the scheduler If a problem is detected you may wish to increase the LOGLEVEL value to get more details However doing so will cause the logs to roll faster and will also cause a lot of possibly unrelated information to clutter up the logs Also be aware of the fact that high LOGLEVEL values will result in large volumes of possibly unnecessary file I O to occur on the scheduling machine Consequently it is not recommended that high LOGLEVEL values be used unless tracking a problem or similar circumstances warrant the I O cost NOTE If high log levels are desired for an extended period of time and your Maui home directory is located on a network filesystem performance may be improved by moving your log directory to a local file system using the LOGDIR parameter A final log related parameter is LOGFACILITY This parameter can be used to focus logging on a subset of scheduler activities This parameter is specified as a list of one or more scheduling facilities as listed in the parameters documentation The logging that occurs is of five major types subroutine information status information scheduler warnings scheduler alerts and scheduler errors These are described in detail below 1 Subroutine Information Each subroutine is logged along with all printable parameters Major subroutines are logged at lower LOGLEVELs while
120. also not a matter of getting it right Cluster resources evolve the workload evolves and even site policies evolve resulting in changing priority needs over time Anecdotal evidence indicates that most sites establish a relatively stable priority policy within a few iterations and make only occasional adjustments to priority weights from that point on Lets look at one more example A site wants to do the following favor jobs in the low medium and high QOS s so they will run in QOS order balance job expansion factor use job queue time to prevent jobs from starving The sample maui cfg is listed below QOSWEIGHT il XFACTORWEIGHT 1 QUBEUETIMEWEIGHT 10 TARGE TQUBUET IMEWEIGHT ali OOSCFG low PRIORITY 1000 QOSCFG medium PRIORITY 10000 QOSCFG high PRIORITY 10000 USERCFG DEFAULT QTTARGET 4 00 00 This example is a bit more complicated but is more typical of the needs of many sites The desired QOS weightings are established by enabling the QOS subfactor using the QOSWEIGHT parameter while the various QOS priorities are specified using QOSCFG XFACTORWEIGHT is then set as this subcomponent tends to establish a balanced distribution of expansion factors across all jobs Next the queuetime component is used to gradually raise the priority of all jobs based on the length of time they have been queued Note that in this case QUEUETIMEWEIGHT was explicitly set to 10 overriding its default value of 1 Finally the TARGETQUEUE
121. alue In general use the speed of a base node is determined and assigned a speed of 1 0 A node that is 50 faster would be assigned a value of 1 5 while a slower node may receive a value which is proportionally less than 1 0 Node speeds do not have to be directly proportional to processor speeds and may take into account factors such as memory size or networking interface Generally node speed information is used to determine proper wallclock limit and CPU time scaling adjustments Node speed information is specified as a unitless floating point ratio and can be specified through the resource manager or with the NODECFG parameter The SPEED specification must be in the range of 0 01 to 100 0 FEATURES Not all resource managers allow specification of opaque node features For these systems the NODECFG parameter can be used to directly assign a list of node features to individual nodes Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 12 3 Node Specific Policies Specification of node policies is fairly limited within Maui mainly because the demand for such policies is limited These policies allow a site to specify on a node by node basis what the node will and will not support Node policies may be applied to specific nodes or applied system wide using the specification NODECFG DEFAULT Note that these policies were introduced over time so not all policies are supported in all versions
122. ameter Or is available in Maui 3 0 7 and Maui will select the resource set containing the fastest higher See Node Set nodes available Overview specifies the tolerance for selection of mixed processor speed nodes A tolerance of X allows a range of processors to be selected subject to the constraint NODESETATTRIBUTE PROCSPEED Speed Max Speed Min NODESETTOLERANCE 0 5 NODESETTOLERANCE lt FLOAT gt 0 0 Exact match only Speed Min lt X NOTE Tolerances are only applicable when NODESETFEATURE is set to PROCSPEED This parameter is available in Maui 3 0 7 and higher See Node Set Overview Maui will only allocate nodes with up to a 50 procspeed difference specifies the length of time after which Maui will sync up a node s expected state with an unexpected reported state IMPORTANT NOTE Maui will not start new jobs on a node as displayed by the showgrid or each matrix output will display data in rows for jobs NODESYNCTIME DD HH MM SS 00 10 00 with an expected state which NODESYNCTIME 1 00 00 does not match the state reported by the resource manager NOTE this parameter is named NODESYNCDEADLINE in Maui 3 0 5 and earlier specifies the weight which will be applied to a job s requested node count before this value is added to the job s cumulative priority NOTE this weight currently only applies when a NODEW
123. and above Many sites possess workloads of varying importance While it may be critical that some jobs obtain resources immediately other jobs are less turnaround time sensitive but have an insatiable hunger for compute cycles consuming every available cycle for years on end These latter jobs often have turnaround times on the order of weeks or months The concept of cycle stealing popularized by systems such as Condor handles such situations well and enables systems to run low priority preemptible jobs whenever something more pressing is not running These other systems are often employed on compute farms of desktops where the jobs must vacate anytime interactive system use is detected Maui s QoS based preemption system allows a dedicated non interactive cluster to be used in much the same way Certain QoS s may be marked with the flag PREEMPTOR others with the flag PREEMPTEE With this configuration low priority preemptee jobs can be started whenever idle resources are available These jobs will be allowed to run until a preemptor job arrives at which point the preemptee job will be checkpointed if possible and vacated This allows near immediate resource access for the preemptor job Using this approach a cluster can maintain near 100 system utilization while still delivering excellent turnaround time to the jobs of greatest value Use of the preemption system need not be limited to controlling low priority jobs Other uses include
124. any number of consumable resources on a per node and per jobs basis Work is under way to allow floating per system resources to be handled as well When a job is started on a set of nodes Maui tracks how much of each available resource must be dedicated to the tasks of the job This allows Maui to prevent per node oversubscription of any resource be it CPU memory swap local disk etc Recent enhancements to Loadleveler version 2 2 and above finally provide a resource manager capable of exercising this long latent capability These changes allow a user to specify per task consumable resources and per node available resources For example a job may be submitted requiring 20 tasks with 2 CPUs and 256 MB per task Thus Maui would allow a node with 1 GB of Memory and 16 processors to allow run 4 of these tasks because 4 tasks would consume all of the available memory Consumable resources allow more intelligent allocation of resources allowing better management of shared node resources No steps are required to enable this capability simply configure the underlying resource manager to support it and Maui will pick up this configuration Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 10 2 Load Balancing Features Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 11 0 General Job Administration 11 1 Job Holds 11 2 Job Pri
125. arameters documentation will show that the default value of SRPERIOD is DAYS Thus specifying this parameter in the example above was unnecessary It was used only to introduce this parameter and indicate that other options exist beyond daily standing reservations Hopefully the next few examples will further clarify the use of standing reservations while expanding on some of the intricacies surrounding their use Note that the above example did not specify where the needed six tasks were to be located If this information is not specified Maui will attempt to locate the needed resources anywhere it can find them The reservation will essentially float to nodes where the needed resources can be found Let s assume you actually wanted to constrain this reservation to a particular set of resources In this case the parameter SRHOSTLIST can be used to specify which nodes can be used for the reservation The following example will do this SRNAME 0 interactive SRHOSTLIST 0 node003 node004 node005 node011 node012 node052 SRTASKCOUNT 0 6 SRRESOURCES 0 PROCS 1 MEM 512 SRDAYS 0 MON TUE WED THU FRI SRSTARTTIME 0 9200200 SRENDTIME 0 17 00 00 SRCLASSLIST 0 interactive The example is now a bit more complex Note that we added a non contiguous list of nodes where the standing reservation can locate the needed resources using the SRHOSTLIST parameter It is important to note that the fact that there is a one to one mapping
126. arify this concept Reservation A requires 4 tasks Each task is defined as 1 processor and 1 GB of memory Node X has 2 processors and 3 GB of memory available Node Y has 2 processors and 1 GB of memory available Node Z has 2 processors and 2 GB of memory available In attempting to collect the resources needed for the reservation Maui would examine each node in turn Maui finds that Node X can support 2 of the 4 tasks needed by reserving 2 processors and 2 GB of memory leaving 1 GB of memory unreserved Analysis of Node Y shows that it can only support 1 task reserving 1 processor and 1 GB of memory leaving 1 processor unreserved Note that the unreserved memory on Node X cannot be combined with the unreserved processor on Node Y to satisfy the needs of another task because a task requires all resources to be located on the same node Finally analysis finds that node Z can support 2 tasks fully reserving all of its resources Both reservations and jobs use the concept of a task description in specifying how resources should be allocated It is important to note that although a task description is used to allocate resources to a reservation this description does not in any way constrain the use of those resources by a job In the above example a job requesting resources simply sees 4 processors and 4 GB of memory available in reservation A If the job has access to the reserved resources and the resources meet the other requirements of the
127. articularly fairness under Maui may be addressed by any combination of the facilities described in the table below Facility Description Example USERCFG john MAXJOB 3 GROUPCFG DEFAULT MAXPROC 64 GROUPCFG staff MAXPROC 128 Specify limits on exactly what resources can be used at any given instant Thottling Policies allow john to only run 3 jobs at a time Allow the group staff to utilize up to 128 total processors and all other groups to utilize up to 64 processors Specify what is most important to the scheduler Using Service based priority factors can allow a site to balance job turnaround time expansion factor or other scheduling performance metrics SERVWEIGHT 1 QUEUETIMEWEIGHT 10 Job Prioritization cause jobs to increase in priority by 10 points for every minute they remain in the queue USERCFG steve FSTARGET 25 0 FSWEIGHT 1 FSUSERWEIGHT 10 Specify usage targets to limits resource access or adjust priority based on historical resource usage Fairshare enable priority based fairshare and specify a fairshare target for user steve such that his job s will be favored in an attempt to keep his job s utilizing at least 25 0 of delivered compute cycles BANKTYPE QBANK BANKSERVER server sys net enable the QBank allocation management system Within the Specify long term allocation manager project or account credential b
128. as homogeneous they quickly evolve as new generations of compute nodes are integrated into the system Research has shown that this integration while improving scheduling performance due to increased scheduler selection can actually decrease average job efficiency A feature called node sets allows jobs to request sets of common resources without specifying exactly what resources are required Node set policy can be specified globally or on a per job basis and can be based on node processor speed memory network interfaces or locally defined node attributes In addition to their use in forcing jobs onto homogeneous nodes these policies may also be used to guide jobs to one or more types of nodes on which a particular job performs best similar to job preferences available in other systems For example an I O intensive job may run best on a certain range of processor speeds running slower on slower nodes while wasting cycles on faster nodes A job may specify ANY OF PROCSPEED 450 500 650 to request nodes in the range of 450 to 650 MHz Alternatively if a simple procspeed homogeneous node set is desired ONEOF PROCSPEED may be specified On the other hand a communication sensitive job may request a network based node set with the configuration ONEOF NETWORK via myrinet ethernet in which case Maui will first attempt to locate adequate nodes where all nodes contain via network interfaces If such a set cannot be found Maui will look for sets of no
129. ased resource based allocations may be configured These allocations may for example allow project X to utilize up to 100 000 processor hours per quarter provide various QoS sensitive charge rates share allocation access etc Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 6 2 Throttling Policies Maui possesses a number of policies which allow an administrator to control the flow of jobs through the system These throttling policies work as filters allowing or disallowing a job to be considered for scheduling by specifying limits regarding system usage for any given moment These policies may be specified as global or specific constraints specified on a per user group account QOS or class basis 6 2 1 Fairness via Throttling Policies 62 1 1 Basic Fairness Policies 6 2 1 2 Multi Dimension Fairness Policies 62 2 Override Limits 6 2 3 Idle Job Limits 6 2 4 Hard and Soft Limits 6 2 1 Fairness via Throttling Policies Significant improvements in the flexibility of throttling policies were introduced in Maui 3 0 7 Those sites using versions prior to this should consult the Maui 3 0 6 style throttling policy configuration documentation Ata high level Maui allows resource usage limits to be specified for in three primary dimensions 6 2 1 1 Basic Fairness Policies Active Job Limits Constrains the total cumulative resource available to active jobs at a given time
130. ation Default Value NONE Description specifies account specific attributes See the flag overview for a description of legal flag values NOTE Only available in Maui 3 0 7 and higher Example ACCOUNTCFG projectX MAXJOB 50 QDEF highprio up to 50 jobs submitted under the account ID projectx will be allowed to execute simultaneously and will be assigned the QOS highprio by default ACCOUNTFSWEIGHT ACCOUNTWEIGHT ADMIN1 lt INTEGER gt lt INTEGER gt space delimited list of user names root specifies the priority weight to be applied to the account fairshare factor See Fairshare Priority Factor ACCOUNTFSWEIGHT 10 specifies the priority weight to be applied to the specified account priority See Credential Priority Factor ACCOUNTWEIGHT 100 users listed under the parameter ADMINI are allowed to perform any scheduling function They have full control over the scheduler and access to all data The first user listed in the ADMINI user list is considered to be the primary admin and is the ID under which maui must be started and run Valid values include user names or the keyword ALL ADMIN1 mauiuser steve scott jenny all users listed have full access to Maui control commands and maui data Maui must be started by and run under the mauiuser user id since mauiuser is the primary admin ADMIN2 space delimited list of user names
131. available and described in detail below FIRSTAVAILABLE MAXBALANCE FASTEST and LOCAL Additional load allocation polices such as may be enabled through extension libraries such as G2 Documentation for the extension library of interest should be consulted 5 2 1 Node Allocation Overview 5 2 2 Resource Based Algorithms 5 2 2 1 CPULOAD 5 2 2 2 FIRSTAVAILABLE 5 2 2 3 LASTAVAILABLE 5 2 2 4 MACHINEPRIO 5 2 2 5 MINRESOURCE 5 2 2 6 CONTIGUOUS 5 2 2 7 MAXBALANCE 5 2 2 8 FASTEST YA DBR A YER ED 5 2 2 9 LOCAL 5 2 3 Time Based Algorithms 5 2 4 Locally Defined Algorithms 5 2 1 Node Allocation Overview Node allocation is important in the following situations heterogeneous system If the available compute resources have differing configurations and a subset of the submitted jobs cannot run on all of the nodes then allocation decisions can significantly affect scheduling performance For example a system may be comprised of two nodes A and B which are identical in all respects except for RAM possessing 256MB and 1GB of RAM respectively Two single processor jobs X and Y are submitted one requesting at least 512 MB of RAM the other at least 128 MB The scheduler could run job X on node A in which case job Y would be blocked until job X completes A more intelligent approach may be to allocate node B to job X because it has the fewest available resources yet still meets the constraints This is somewhat of a
132. avior would require two steps First the user credential subcomponent would need to be enabled and second john would need to have his relative priority specified Take a look at the example maui cfg USERWEIGHT 1 USERCFG john PRIORITY 300 The USER priority subcomponent was enabled by setting the USERWEIGHT parameter In fact the parameters used to specify the weights of all components and subcomponents follow this same WEIGHT naming convention i e RESWEIGHT TARGETQUEUETIMEWEIGHT etc The second part of the example involved specifying the actual user priority for the user john This was accomplished using the USERCFG parameter Why was the priority 300 selected and not some other value Is this value arbitrary As in any priority system actual priority values are meaningless only relative values are important In this case we are required to balance user priorities with the default queue time based priorities Since queuetime priority is measured in minutes queued see table above the user priority of 300 will make a job by user john on par with a job submitted 5 minutes earlier by another user Is this what the site wants Maybe maybe not The honest truth is that most sites are not completely certain what they want in prioritization at the onset Most often prioritization is a tuning process where an initial stab is made and adjustments are then made over time Unless you are an exceptionally stable site prioritization is
133. b the tools subdirectory specifies the weight assigned to FSACCOUNTWEIGHT lt INTEGER gt 0 the account subcomponent of jc aCCOUNTWEIGHT 10 the fairshare component of priority Poo FSCAP 10 0 specifies the maximum allowed value for a job s total Maui will not allow a job s pre weighted fairshare FSCAP lt DOUBLE gt 0 NO CAP pre weighted fairshare component to exceed 10 0 component ie Priority FSWEIGHT MIN FSCAP FSFACTOR F SCONFIGFILE STRING gt fs cfg FSDECAY KDOUBLE gt s Cs SC 1 0 FSDEPTH me 7 FSGROUPWEIGHT lt INTEGER gt 0 FSGROUPWEIGHT 4 specifies the length of each FSINTERVAL 12 00 00 FSINTERVAL DD HH MM SS salma fairshare window track fairshare usage in 12 hour blocks specifies the unit of tracking fairshare usage FSPOLICY DEDICATEDPES Scie one of the following DEDICATEDPS ou AA aes DEDICATEDPES edicated processor seconds Maui will track fairshare usage by dedicated DEDICATEDPES tracks process equivalent seconds dedicated processor equivalent seconds specifies the priority weight FSQOSWEIGHT lt INTEGER gt 0 assigned to the QOS fairshare subcomponent specifies the priority weight FSUSERWEIGHT lt INTEGER gt 0 assigned to the user fairshare FSUSERWEIGHT 8 subfactor specifies the priority weight FSWEIGHT lt INTEGER gt 0 assigned to the summation of FSWEIGHT 500 the fai
134. behavior slightly and allow the reservation to allocate resources even if reservation conflicts exist Other standing reservation parameters not covered here include SRPARTITION and SRCHARGEACCOUNT These parameters are described in some detail in the Maui parameters documentation 7 1 5 3 Configuring Administrative Reservations A default reservation with no ACL is termed a SYSTEM reservation It blocks access to all jobs because it possesses an empty access control list It is often useful when performing administrative tasks but cannot be used for enforcing resource usage policies Under construction See Also N A Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 7 2 Partitions Partitions are a logical construct which divide available resources By default a given job may only utilize resources within a single partition and any resource i e compute node may only be associated with a single partition In general partitions are organized along physical or political boundaries For example a cluster may consist of 256 nodes containing four 64 port switches This cluster may receive excellent interprocess communication speeds for parallel job tasks located within the same switch but sub stellar performance for tasks which span switches To handle this the site may choose to create four partitions allowing jobs to run within any of the four partitions but not span them While partiti
135. bestfit approach in the configured resource dimension and is essentially what is done by the MINRESOURCE algorithm shared node system Shared node systems are most often involve SMP nodes although this is not mandatory Regardless when sharing the resources of a given node amongst tasks from more than one job resource contention and fragmentation issues arise Most current systems still do not do a very good job of logically partitioning the resources i e CPU Memory network bandwidth etc available on a given node Consequently contention often arises between tasks of independent jobs on the node This can result in a slowdown for all jobs involved which can have significant ramifications if large way parallel jobs are involved On large way SMP systems i e gt 32 processors node job packing can result in intra node fragmentation For example again take two nodes A and B each with 64 processors Assume they are currently loaded with various jobs and have 24 and 12 processors free respectively Two jobs are submitted Job X requesting 10 processors and job Y requesting 20 processors Job X can start on either node but starting it on node A will prevent job Y from running An algorithm to handle intra node fragmentation is pretty straightforward for a single resource case but what happens when jobs request a combination of processors memory and local disk Determining the correct node suddenly gets significantly more difficu
136. bs In this example all holds were released from these two jobs Related Commands You can place a hold on a job using the set hold command Notes None Copyright 1998 Maui High Performance Computing Center All rights reserved Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ releaseres releaseres ARGUMENTS lt RESERVATION ID gt lt RESERVATION ID ARGUMENTS h USAGE HELP Purpose Release existing reservation Access Users can use this command to release any reservation they own Level 1 and level 2 Maui administrators may use this command to release any reservation This command can be run by any user Parameters RESERVATION ID Name of reservation to release Flags h Help for this command Description This command allows Maui Scheduler Administrators to release any user group account job or system reservation Users are allowed to release reservations on jobs they own Note that releasing a reservation on an active job has no effect since the reservation will be automatically recreated Example Release two existing reservations Q releaseres system 1 bob 2 released User reservation system 1 released User reservation bob 2 Related Commands You can view reservations with the showres command You can set a reservation using the set res command Notes See the Reservation document for more information Copyright
137. ces into native resource manager objects attributes and commands Information on creation a new scheduler resource manager interface can be found in the Adding New Resource Manager Interfaces section 13 1 1 1 Resource Manager Commands In the simplest configuration Maui interacts with the resource manager using the four primary functions listed below GETJOBINFO Collect detailed state and requirement information about idle running and recently completed jobs GETNODEINFO Collect detailed state information about idle busy and defined nodes STARTJOB Immediately start a specific job on a particular set of nodes CANCELJOB Immediately cancel a specific job regardless of job state Using these four simple commands Maui enables nearly its entire suite of scheduling functions More detailed information about resource manager specific requirements and semantics for each of these commands can be found in the specific resource manager overviews LL PBS or WIKI In addition to these base commands other commands are required to support advanced features such a dynamic job support suspend resume gang scheduling and scheduler initiated checkpoint restart 13 1 1 2 Resource Manager Flow Early versions of Maui i e Maui 3 0 x interacted with resource managers in a very basic manner stepping through a serial sequence of steps each scheduling iteration These steps are outlined below 1 load global resource information
138. ch priority subcomponent is calculated as lt COMPONENT WEIGHT gt lt SUBCOMPONENT WEIGHT gt lt PRIORITY SUBCOMPONENT VALUE gt Each priority component contains one or more subcomponents as described in the Priority Component Overview For example the Resource component consists of Node Processor Memory Swap Disk and PE subcomponents While there are numerous priority components and many more subcomponents a site need only focus on and configure the subset of components related to their particular priority needs In actual usage few sites use more than a small fraction usually 5 or less of the available priority subcomponents This results in fairly straightforward priority configurations and tuning By mixing and matching priority weights sites may generally obtain the desired job start behavior At any time the diagnose p command can be issued to determine the impact of the current priority weight settings on idle jobs Likewise the command showegrid can assist the admin in evaluating priority effectiveness on historical system usage metrics such as queue time or expansion factor As mentioned above a job s priority is the weighted sum of its activated subcomponents By default the value of all component and subcomponent weights is set to 1 and 0 respectively The one exception is the QUEUETIME subcomponent weight which is set to 1 This results in a total job priority equal to the period of time the job has been queued causi
139. cheduler Defer Hold is in place a temporary hold used when a job has been unable to start after a specified number of attempts This hold is automatically removed after a short period of time Job is in the LoadLeveler state NQ indicating the job s controlling scheduling daemon in unavailable A summary of the job queue s status is provided at the end of the output Example 2 o fr28n fr28n0 fri7n0 fr28n1 fr28n0 fr28n0 fr28n fr28nl fri7n0 fr28n0 fr28n EITO frl7nl fr28n0 fr28n0 frl7nl fr1l7n0 frl7nl fr28n0 fr28nl 23 Jobs frl7n09 529 fr28n13 701 f r28n13 631 showq r JobName n r w 13 109 Pe23032 8 1349 5 4355 5 2098 5 2095 13 683 5 4354 8 1341 S209 7 13 682 5 8 1328 0 1467 7 2300 1 1851 0 1466 21395 1 1388 7 2304 1 3091 SCQaoomrooouoeooueooem ooo OOO PPNNFPWREEFEFPRPWWWENWREEPRPWENE ANHANHDAADAHAAAAAADAADHAAAAAADAD 99 98 9T 98 94 98 99 98 99 99 99 99 98 OF 99 3 98 99 98 99 95 63 97 98 Effic 37 57 94 91 26 56 75 90 67 70 83 69 12 60 10 01 51 9I 89 29 46 62 87 XFactor rr a H PNHRRRPERRPRRER EF Et DOOBROONNWOOFOUFDOONDTOWODOOO 251 of 254 Processors Active The fields are as follows JobName S Pa Effic XFactor Q User Group Nodes Name of running job Job State Either R for Running or S for Starting Partition in which job is running
140. cies IGNSYSTEM jobs should ignore all system throttling policies IGNALL jobs should ignore all user group and account throttling policies Example QOSCFG express FLAGS IGNSYSTEM 7 3 3 Managing QoS Access Managing which jobs can access which privileges is handled via the QOSCFG parameter Specifically this parameter allows the specification of a access control list based on a job s user group account and queue credentials To enable QoS access the QLIST and or QDEF attributes of the appropriate user group account or queue should be specified using the parameters USERCFG GROUPCFG ACCOUNTCEG and CLASSCFG respectively Example USERCFG john QDEF geo QLIST geo chem staff GROUPCFG systems QDEF development CLASSCFG batch QDEF normal See also N A Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 8 0 Optimizing Scheduling Behavior Backfill Node Sets and Preemption 8 2 Backfill 8 3 Node Sets H 8 4 Preemption Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 8 1 Optimization Overview Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 8 2 Backfill 391 Backfill Overview 8 2 2 Backfill Algorithm 8 2 3 Configuring Backfill 8 2 1 Backfill Overview Backfill is a scheduling optimization which allows a scheduler to make better use of avail
141. ckfilled Includes summary of job weighted backfill job percent and total samples BFNHRUN Number of node hours backfilled Includes summary of job weighted backfill node second percentage and total samples JOBCOUNT Number of jobs Includes summary of total jobs and total samples JOBEFFICIENCY Job efficiency Includes summary of job weighted job efficiency percent and total samples MAXBYPASS Maximum bypass count Includes summary of overall maximum bypass and total samples MAXXFACTOR Maximum expansion factor Includes summary of overall maximum expansion factor and total samples NHREQUEST Node hours requested Includes summary of total node hours requested and total samples NHRUN Node hours run Includes summary of total node hours run and total samples QOSDELIVERED Quality of service delivered Includes summary of job weighted quality of service success rate and total samples WCACCURACY Wall clock accuracy Includes summary of overall wall clock accuracy and total samples NOTE The STATISTICTYPE parameter value must be entered in uppercase characters Flags h Help for this command Description This command displays a table of the selected Maui Scheduler statistics such as expansion factor bypass count jobs node hours wall clock accuracy and backfill information Example showgrid AVGXFACTOR Average XFactor Grid NODES 00 02 00 00 04 00 00 08 00 00 16 00 00 32 00 01 04 00 02 08 00 04 16 00 08 32 0
142. clock limit Using this equation it can be seen that short running jobs will have an xfactor that will grow much faster over time than the xfactor associated with long running jobs The table below demonstrates this favoring of short running jobs Job Queue Time 1 hour 2 hours 4 hours 8 hours 16 hours XFactor for 1 hour 1 1 1 1 1 4 1 1 86 D 1 6 1 1 job 2 00 3 00 5 00 9 00 17 0 XFactor for 4hour 1 4 4 2 4 4 44 4 4 84 4 4 6 4 4 job 1 25 1 50 2 00 3 00 5 0 Since XFactor is calculated as a ratio of two values it is possible for this subcomponent to be almost arbitrarily large potentially swamping the value of other priority subcomponents This can be addressed either by using the subcomponent cap XFCAP or by using the XFMINWCLIMIT parameter If the later is used the calculation for the xfactor subcomponent value becomes XFACTOR 1 lt EFFQUEUETIME gt MAX lt XFMINWCLIMIT gt lt WALLCLOCKLIMIT gt The use of the XFMINWCLIMIT parameter allows a site to prevent very short jobs from causing the Xfactor subcomponent to grow inordinately Some sites consider XFactor to be a more fair scheduling performance metric than queue time At these sites job XFactor is given far more weight than job queue time when calculating job priority and consequently job XFactor distribution tends to be fairly level across a wide range of job durations i e A flat XFactor distribution of 1 0 would result in a
143. covers They need an extensive set of buttons and knobs to both enable management enforced policies and tune the system to obtain desired Statistics 1 1 Value of a Batch System 1 2 Philosophy and Goals of the Maui Scheduler Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 1 1 Value of a Batch System Batch systems provide a mechanism for submitting launching and tracking jobs on a shared resource These services fullfil one of the major responsibilities of a batch system providing centralized access to distributed resources This greatly simplifies the use of the cluster s distributed resources allowing users a single system image in terms of the management of their jobs and the aggregate compute resources available However batch systems must do much more than provide a global view of the cluster As with many shared systems complexities arise when attempting to utilize compute resources in a fair and effective manner These complexities can lead to poor performance and significant inequalities in usage With a batch system a scheduler is assigned the job of determining when where and how jobs are run so as to maximize the output of the cluster These decisions are broken into three primary areas 1 1 1 Traffic Control 113 Optimizations 1 1 1 Traffic Control A scheduler is responsible for preventing jobs from interfering with each other If jobs are allowed to contend
144. cular node or partition will be assigned these default attribute values See the Partition Overview for more information Direct maui parameter specification Maui also provides a parameter named NODECFG which allows direct specification of virtually all node attributes supported via other mechanisms and also provides a number of additional attributes not found elsewhere For example a site may wish to specify something like the following NODECFG node031 MAXJOB 2 PROCSPEED 600 PARTITION small These approaches may be mixed and matched according to the site s local needs Precedence for the approaches generally follows the order listed above in cases where conflicting node configuration information is specified through one or more mechanisms 12 1 Node Location Partitions Frames Queues etc 12 2 Node Attributes Node Features Speed etc 12 3 Node Specific Policies MaxJobPerNode etc Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 12 1 Node Location Nodes can be assigned three types of location information based on partitions frames and or queues 12 1 1 Partitions 12 1 2 Frames 12 1 3 Queues 12 1 3 1 OpenPBS Queue to Node Mapping 12 1 1 Partitions The first form of location assignment the partition allows nodes to be grouped according to physical resource constraints or policy needs By default jobs are not allowed to span more than one partition
145. d due to resource manager corruption RESOURCECAP 0 1000 specifies the maximum allowed The total resource priority factor component of a job s RESOURCECAP X lt DOUBLE gt 0 NO CAP pre weighted job resource priority will not be allowed to exceed 1000 i e priority factor Priority RESWEIGHT MIN RESOURCECAP lt RESOURCEFACTOR gt ase lt POLICY gt lt RESOURCETPYE gt SeennecaRL ABT ET once where 7 o DEDICATED PROCS COMBINED MEM POLICY is one of COMBINED specifies how Maui will ee en a RESOURCEAVAILABILITYPOLICY DEDICATED or UTILIZED xxxxx COMBINED evaluate resource availability on Maui will ignore resource utilization information in gt a per resource basis locating available processors for jobs but will use both ann dedicated and utilized information i RESOURCETYPE is one of edicated and utilized memory information in PROC MEM SWAP or DISK determining memory availability lt POLICY gt lt ACTION gt lt RESOURCE gt lt RESOURCE gt where POLICY is one of ALWAYS EXTENDEDVIOLATION or BLOCKEDWORKLOADONLY specifies how the scheduler should handle job which utilize RESOURCELIMITPOLICY ALWAYS CANCEL MEM RESOURCELIMITPOLICY no limit enforcement more resources than they Maui will cancel all jobs which exceed their requested where ACTION is one of CANCEL request NOTE Onl
146. d priority of a job to meet current needs Current needs often are broken into two categories A The need to run an admin test job as soon as possible B The need to pacify an irate user Under Maui the setspri command can be used to handle these issues in one of two ways This command allows the specification of either a relative priority adjustment or the specification of a absolute priority Using absolute priority specification administrators can set a job priority which is guaranteed to be higher than any calculated value Where Maui calculated job priorities are in the range of 0 to 1 billion system admin assigned absolute priorities start at 1 billion and go up Issuing the command setspri lt PRIO gt lt JOBID gt for example will assign a priority of 1 billion lt PRIO gt to the job Thus setspri 5 job 1294 with set the priority of job job 1294 to 1000000005 Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 5 2 Node Allocation While job prioritization allows a site to determine which job to run node allocation policies allow a site to specify how available resources should be allocated to each job The algorithm used is specified by the parameter NODEALLOCATIONPOLICY There are multiple node allocation policies to choose from allowing selection based on reservation constraints node configuration available resource constraints and other issues The following algorithms are
147. d significant priority tuning is unnecessary after a few days Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved A 2 Case Study Semi Partitioned Heterogeneous Cluster Dedicated to Parallel and Time sharing Serial Jobs Overview A site possessing a mixture of uniprocessor and dual processor nodes desires to dedicate a subset of nodes to time sharing serial jobs a subset to parallel batch jobs and provide a set of nodes to be used as overflow Resources Compute Nodes Group A 16 uniprocessor Linux based nodes each with 128 MB of RAM and 1 GB local scratch space Group B 8 2 way SMP Linux based nodes each with 256 MB of RAM and 4 GB local scratch space Group C 8 uniprocessor Linux based nodes each with 192 MB of RAM and 2 GB local scratch space Resource Manager OpenPBS 2 3 Network 100 MB switched ethernet Workload Job Size range in size from 1 to 32 processors with approximately the following quartile job frequency distribution 1 2 3 8 9 24 and 25 32 nodes Job Length jobs range in length from 1 to 24 hours Job Owners job are submitted from 6 major groups consisting of a total of about 50 users NOTES During prime time hours the majority of jobs submitted are smaller short running development jobs where users are testing out new code and new data sets The owners of these jobs are often unable to proceed with their work until a job they have submitted comple
148. des containing the other specified network interfaces In highly heterogeneous clusters the use of node sets have been found to improve job throughput by 10 to 15 Node sets can be requested on a system wide or per job basis System wide configuration is accomplished via the NODESET parameters while per job specification occurs via the resource manager extensions In all cases node sets are a dynamic construct created on a per job basis and built only of nodes which meet all of the jobs requirements As an example let s assume a large site possessed a Myrinet based interconnect and wished to whenever possible allocate nodes within Myrinet switch boundaries To accomplish this they could assign node attributes to each node indicating which switch it was associated with ie switchA switchB etc and then use the following system wide node set configuration NODESETPOLICY ONEOF NODESETATTRIBUTE FEATURE NODESETDELAY 0 00 00 NODESETLIST SwitchA switchB switchC switchD The NODESETPOLICY parameter tells Maui to allocate nodes within a single attribute set Setting NODESETATTRIBUTE to FEATURE specifies that the node sets are to be constructed along node feature boundaries The next parameter NODESETDELAY indicates that Maui should not delay the start time of a job if the desired node set is not available but adequate idle resources exist outside of the set Setting this parameter to zero basically tells Maui to attempt to use a node set i
149. e Attribute Value pairs h display command usage help check job start eligibility subject to specified throttling policy level 1 lt POLICYLEVEL gt 501 ICYLEVEL gt can be one of HARD SOFT or OFF r lt RESID gt check job access to specified reservation V display verbose job state and eligibility information Description This command allows any Maui administrator to check the detailed status and resources requirements of a job Additionally this command performs numerous diagnostic checks and determines if and where the could potentially run Diagnostic checks include policy violations See the Throttling Policy Overview for details reservation constraints and job to resource mapping If a job cannot run a text reason is provided along with a summary of how many nodes are and are not available If the v flag is specified a node by node summary of resource availability will be displayed for idle jobs If a job cannot run one of the following reasons will be given Reason Description job has hold in place one or more job holds are currently in place linsufficient idle procs adequate idle processors are available but these do not idle procs do not meet requirements meet job requirements job has specified a minimum start date which is still in start date not reached the future expected state is not idle job is in an unexpected state stateisnotide job is not in the idle state dependency is not met
150. e in which the highest priority batch job cannot run Maui can make a reservation for this highest priority job but because their are no constraints on the background load Maui cannot determine when this background load will drop enough to allow this job to run By default it optimistically attempts a reservation for the next scheduling iteration perhaps 1 minute out The problem is that this reservation now exists one minute out and when Maui attempts to backfill it can only consider jobs which request less than one minute or which can fit beside this high priority job The next iteration Maui still cannot run the job because the background load has not dropped and again creates a new reservation for one minute out The background load has basically turned batch scheduling into an exercise in resource scavenging If the priority job reservation were not there other smaller queued jobs might be able to run Thus to maximize the scavenging effect the scheduler should be configured to allow this high priority job first dibs on all available resources but prevent it from reserving these resources if it cannot run immediately Configuration The configuration needs to accomplish several main objectives including track the background load to prevent oversubscription favor small short jobs to maximize job turnaround prevent blocked high priority jobs from creating reservations interface to an allocation manager to charge fo
151. e limited Example gt showstart job001 job Job001 requires 2 procs for 0 33 20 Earliest start is in 1 40 00 on Thu Jan 1 01 16 40 Earliest Completion is in 2 13 20 on Thu Jan 1 01 50 00 Best Partition DEFAULT Related Commands checkjob showres Notes Since the information provided by this job is only highly accurate if the job is highest priority or if the job has a reservation sites wishing to make decisions based on this information may want to consider using the RESERVATIONDEPTH parameter to increase the number of priority based reservations This can be set so that most or even all idle jobs receive priority reservations and make the results of this command generally useful The only caution of this approach is that increasing the RESERVATIONDEPTH parameter more tightly constrains the decisions of the scheduler and may resulting in slightly lower system utilization typically less than 8 reduction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved L showstate showstate h Purpose Summarizes the state of the system Permissions This command can be run by any Maui Scheduler Administrator Parameters None Flags h Shows help for this command Description This command provides a summary of the state of the system It displays a list of all active jobs and a text based map of the status of all nodes and the jobs they are servicing Simple diagnostic tests are also pe
152. e table below presents the primary status commands and flags The Command Overview lists all available commands Command Flags Description cialiak display job state resource requirements environment constraints secon credentials history allocated resources and resource utilization dhean display node state resources attributes reservations history and statistics diagnose E j display summarized job information and any unexpected state diagnose nose n display summarized node information and any unexpected state display summarized node information and any unexpected state node information and any unexpected state showgrid display various aspects of scheduling performance across a job duration job size matrix display various views of currently queued active idle and non eligible howgq r l show rl i ob Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved ___ 4 3 Job Management Commands Maui shares job management tasks with the resource manager Typically the scheduler only modifies scheduling relevant aspects of the job such as partition access job priority charge account hold state etc The table below covers the available job management commands The Command Overview lists all available commands Command Flags Description canceljob cancel existing job eheeiiah display job state resource requirements environment constraints
153. ecify node type via node Maui will interpret all node features with the leading FEATURENODETYPEHEADER lt STRING gt NONE features ie LL features or PBS string xnt as a nodetype specification as used by node attributes QBank and other allocation managers and assign the associated value to the node i e xntFast specifies the header used to FRATURFEARTITTONHRADER xpt FEATUREPARTITIONHEADER lt STRING gt NONE specify node partition via node Maui will interpret all node features with the leading features ie LL features or PBS string xpt as a partition specification and assign the node attributes associated value to the node i e xptGold specifies the header used to extract node processor speed via node features i e LL features or PBS node attributes NOTE FEATUREPROCSPEEDHEADER xps FEATUREPROCSPEEDHEADER lt STRING gt NONE Adding a trailing character Maui will interpret all node features with the leading will specifies that only features string xps as a processor speed specification and with a trailing number be assign the associated value to the node i e xps950 interpreted For example the header sp will match sp450 but not sport specifies the name of the program to be run at the FEEDBACKPROGRAM var maui fb pl completion of each job If not FEEDBACKPROGRAM lt STRING gt NONE fully qualified Maui will Maui will run the specified program at the completion attempt to locate this program in Of each jo
154. ed within the lt X gt Interface c module This process is quite easy and involves merely extending existing resource manager specific case statements within the general resource manager calls The vast majority of the development effort in entailed in creating the resource manager specific data collection and job management calls These calls populate Maui data structures and are responsible for passing Maui scheduling commands on to the resource manager The base commands are GetJobs GetNodes StartJob and CancelJob but if the resource manager support is available extended functionality can be enabled by creating commands to suspend resume jobs checkpoint restart jobs and or allow support of dynamic jobs If the resource manager provides a form of event driven scheduling interface this will also need to be enabled The PBSInterface c module provides a template for enabling such an interface within the PBSProcessEvent call 13 4 2 Wiki Interface The Wiki interface is a good alternative if the resource manager does not already support some form of existing scheduling API For the most part use of this API requires the same amount of effort as creating a resource manager specific interface but development effort focussed within the resource manager Since Wiki is already defined as a resource manager type no modifications are required within Maui Additionally no resource manager specific library or header file is required However wi
155. ee Group GROUPCEG account specific priority SEE ACCOUNTCEG QOS QOS specific priority See QOSCFG QOS specific priority See QOSCFG See QOSCFG class queue specific priority See a CLASSCEG iB user based historical usage See fairshare usage FSUSER FSUSER per fase Overview FSGROUP group based historical usage See Fairshare Overview account based historical usage See Eo OUNE Fairshare Overview QOS base historical usage See Fairshare ESQ Overview FSCLASS class queue based historical usage See Fairshare Overview NODE number of nodes requested PROC pumber of processors requested MEM total real memory requested in MB SWAP total virtual memory requested in MB DISK__ ftotal local disk requested in MB IPS total proc seconds requested PE total processor equivalent requested WALLTIME ftotal walltime requested in seconds es ened QUEUETIME time job has been queued in minutes XFACTOR minimum job expansion factor number of times job has been bypassed Bypass by backfill TARGET TARGETQUEUETIME time until queuetime target is reached target service levels exponential TARGETXEACTOR distance to target expansion factor exponential USAGE consumed resources CONSUMED proc seconds dedicated to date active jobs only REMAINING proc seconds outstanding PERCENT percent of required walltime consumed 5 1 2 1 Credential CRED Component The credential co
156. eferral NOTE As of Maui 3 0 7 the reason a job is deferred or placed in a batch hold is stored in memory but is not checkpointed Thus this info is available only until Maui is recycled at which point the checkjob command will no longer display this reason info under construction Controlling Backfill Reservation Behavior Reservation Thresholds Reservation Depth Resource Allocation Method First Available Min Resource Last Available WallClock Limit Allowing jobs to exceed wallclock limit MAXJOBOVERRUN Using Machine Speed for WallClock limit scaling USEMACHINESPEED Controlling Node Access NODEACCESSPOLICY Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 11 2 Job Priority Management Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 11 3 Suspend Resume Handling Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 11 4 Checkpoint Restart Facilities Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 12 0 General Node Administration Since Maui interoperates with a number of resource managers of varying capabilities it must possess a somewhat redundant set of mechanisms for specifying node attribute location and policy information Maui determines a node s configuration through one or more o
157. efined and then associated with a subset of available compute resources With such systems such as Loadleveler or PBSPro these queue to node mappings are automatically detected On resource managers which do not provide this service Maui provides alternative mechanisms for enabling this feature 12 1 3 1 OpenPBS Queue to Node Mapping Under OpenPBS queue to node mapping can be accomplished setting the queue acl_hosts parameter to the mapping hostlist desired within PBS Further the acl_host_enable parameter should be set to False NOTE Setting acl_hosts and then setting acl_host_enable to True will constrain the list of hosts from which jobs may be submitted to the queue Prior to Maui 3 0 7p3 queue to node mapping was only enabled when acl_host_enable was set to True thus for these versions the acl_host list should always include all submission hosts Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 12 2 Node Attributes Nodes can possess a large number of attributes describing their configuration The majority of these attributes such as operating system or configured network interfaces can only be specified by the direct resource manager interface However the number and detail of node attributes varies widely from resource manager to resource manager Sites often have interest in making scheduling decisions based on scheduling attributes not directly supplied by the resource manager Config
158. elds marked with an asterisk are only displayed when set or when the v flag is specified Examples Example 1 gt checkjob v job05 checking job job05 State Idle User john Group staff Account NONE WallTime 0 00 00 Limit 6 00 00 Submission Time Mon Mar 2 06 34 04 Total Tasks 2 Req 0 TaskCount 2 Partition ALL Network hps_user Memory gt 0 Disk gt 0 Features NONE Opsys AIX43 Arch R6000 Class batch 1 ExecSize 0 ImageSize 0 Dedicated Resources Per Task Procs 1 NodeCount 0 IWD NONE Executable cmd QOS DEFAULT Bypass 0 StartCount 0 Partition Mask ALL Holds Batch batch hold reason Admin PE 2 00 StartPriority 1 job cannot run job has hold in place job cannot run insufficient idle procs 0 available Note that the example job cannot be started for two different reasons e It has a batch hold in place There are no idle resources currently available See also diagnose j display additional detailed information regarding jobs Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ checknode checknode NODE h Purpose Displays state information and statistics for the specified node Permissions This command can be run by any Maui Scheduler Administrator Parameters NODE Node name you want to check Flags h Help for this command Description This command shows detai
159. enter showbf m 256 to request nodes with 256 MB memory n Show backfill information for a specified number of nodes That is this flag can be used to force showbf to display only windows larger than a specified size p Show backfill information for the specified partition q Show information for the specified QOS u Show backfill information only for specified user Description This command can be used by any user to find out how many processors are available for immediate use on the system It is anticipated that users will use this information to submit jobs that meet these criteria and thus obtain quick job turnaround times This command incorporates down time reservations and node state information in determining the available backfill window Example 1 showbf backFill window user Jjohn group staff partition ALL Mon Feb 16 08 28 54 partition FAST 9 procs available for 4 54 18 partition SLOW 34 procs available for 10 25 30 26 procs available for 7 00 19 1 proc available with no timelimit In this example a job requiring up to 34 processors could be submitted for immediate execution in partition 2 as long as it required less than 10 hours 25 minutes Likewise jobs requiring up to 26 processors that complete in less than 7 hours could also run in partition SLOW A single processor job with arbitrary wallclock limits could also run in this partition
160. er Construction Overview An 8 node 32 processor heterogeneous SP2 system is to be scheduled in a shared node manner Resources Compute Nodes 8 node 32 processor 24 GB SP2 system Resource Manager Loadleveler Network IBM High Performance Switch essentially All to All connected Workload Job Size range in size from 1 to 16 processors Job Length jobs range in length from 15 minutes to 48 hours Job Owners various Constraints Must do Goals Should do Analysis Configuration Monitoring Conclusions Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ A 5 Case Study Multi Queue Cluster with QOS and Charge Rates Overview A 160 node uniprocessor Linux cluster is to be used to support various organizations within an enterprise The ability to receive improved job turnaround time in exchange for a higher charge rate is required A portion of the system must be reserved for small development jobs at all times Resources Compute Nodes 128 800 MHz uniprocessor nodes w 512 MB each running Linux 2 4 32 1 2 GHz uniprocessor nodes w 2 GB each running Linux 2 4 Resource Manager OpenPBS 2 3 Network 100 MB ethernet Workload Job Size range in size from 1 to 80 processors Job Length jobs range in length from 15 minutes to 24 hours Job Owners various Constraints Must do The management desires the following queue structure Qu
161. ers taking steps to improve their individual scheduling performance results in better job packing lower queue time and better system utilization overall Because this behavior is beneficial to the system at large system admins and management should encourage this behavior and provide the best possible information to them There are two primary types of information which help users make improved decisions Cluster wide resource availability information and per job resource utilization information 15 1 1 Improving Job Size Duration Requests Maui provides a number of informational commands which help users make improved job management decisions based on real time cluster wide resource availability information These commands include showbf showgrid and showq Using these commands a user can determine what resources are available and what job configurations statistically receive the best scheduling performance 15 1 2 Improving Resource Requirement Specification A job s resource requirement specification tells the scheduler what type of compute nodes are required to run the job These requirements may state that a certain amount of memory is required per node or that a node have a minimum processor speed At many sites users will determine the resource requirements needed to run an initial job Then for the next several years they will use the same basic batch command file to run all of their remaining jobs even though the resource requireme
162. ervation regardless of job priority DEDICATED jobs can preempt restartable jobs by essentially requeueing them if this allows the QOS job to start earlier RESTARTPREEMPT job may only utilize resources within accessible reservations If lt RESID gt is specified job may only utilize resources within the specified reservation USERESERVED lt RESID gt Example QOSCFG hiprio FLAGS NOBF PREEMPTEER Example 2 OQOSCFG chem b FLAGS USERESERVED chemistry 7 3 2 3 Policy Exemptions Individual QoS s may be assigned override policies which will set new policy limits regardless of user group account or queue limits Particularly the following policies may be overridden MAXJOB MAXPROC MAXNODE Example QOSCFG staff MAXJOB 48 In addition to overriding policies QoS s may also be used to allow particular jobs to ignore policies by setting the QoS FLAG attribute QOS Flags IGNJOBPERUSER IGNPROCPERUSER IGNPSPERUSER IGNJOBQUEUEDPERUSER IGNJOBPERGROUP IGNPROCPERGROUP IGNPSPERGROUP IGNJOBQUEUEDPERGROUP IGNJOBPERACCOUNT IGNPROCPERACCOUNT IGNPSPERACCOUNT IGNJOBQUEUEDPERACCOUNT IGNS YSMAXPROC IGNSYSMAXTIME IGNSYSMAXPS IGNSRMAXTIME jobs should ignore standing reservation MAXTIME constraints IGNUSER jobs should ignore all user throttling policies IGNGROUP jobs should ignore all group throttling policies IGNACCOUNT jobs should ignore all account throttling poli
163. es resources contained within this listed access to the resources reserved by standing reservation reservation 2 one or more of the following space delimited specifies which days of the SRDAYS 1 Mon Tue Wed Thu Fri SRDAYS X Mon Tue Wed Thu Fri Sat Sun ALL week the standing reservation standing reservation 1 will be active on Monday thru or will be active Friday ALL specifies the number of standing SRDEPTH 1 7 SRDEPTH X lt INTEGER gt 2 reservations which will be specifies that standing reservations will be created for created one per day standing reservation 1 for today and the next 6 days STSTARTTIME 2 8 00 00 specifies the time of day the SRENDTIME 2 17 00 00 SRENDTIME X UHH MM SS 24 00 00 standing reservation becomes inactive standing reservation 2 is active from 8 00 AM until 5 00 PM specifies the required node SRFEATURES 3 wide fddi SRFEATURES X space delimited list of node features NONE features for nodes which will be all nodes used in the standing reservation must have part of the standing reservation both the wide and fddi node attributes colon delimited list of zero or more of the following flags SINGLEUSE S SRFLAGS 1 BYNAME BYNAME specifes special reservation SRFLAGS PREEMPTEE NONE attributes See Managing Jobs may only access the resources within this SLIDEFORWARD Reservations for details reservation if they explicitly request the reservation by FORCE name only enabled in Maui
164. es administration a bit easier SRTASKCOUNT allocated in units called tasks where a task is a collection of resources which must be allocated together on a single node The next parameter SRRESOURCES indicates what resources must be included in each task In this case Maui must locate and reserve 1 processor and 512 MB of memory together on the same node for each task requested SRPERIOD states that this reservation is periodic on a daily basis with the actual days of the week which the standing reservation should be enabled specified using SRDAYS The time of day during which the requested tasks are to be reserved are specified using SRSTARTTIME and SRENDTIME Finally the SRCLASSLIST parameter is used to indicate that jobs requesting the class interactive should be allowed to use this reservation 7 1 5 2 2 Specifying Reservation Resources This is a lot of new information to digest Not all of the parameters are needed in all cases For example by default SRRESOURCES is set to PROCS 1 which indicates that each task should reserve all of the processors on the node on which it is located This in essence creates a one task equals one node mapping In many cases particularly for all uniprocessor systems this default behavior is probably easiest to work with However when SMP systems are thrown into the mix SRRESOURCES provides a powerful means of specifying an exact multi dimensional resource set An examination of the p
165. es and memory Memory Requirement Breakdown portion shows information about the current workload profile In this example the system monitored is a heterogeneous environment consisting of eight 64 MB RAM nodes 144 128 MB nodes etc with a total of 288 nodes The third column indicates the percentage of total nodes that meet this memory criteria For example the eight 64 MB nodes make up 2 78 of the 288 total nodes in the system The idle job queue monitored in this example consists of numerous jobs consisting of a total of 44 381 node hours of work The node hour workload of jobs that have specific node memory requirements are assigned to the corresponding memory class If no specific memory requirement is specified the job s node hours are assigned to the lowest memory class in this case the 64 MB nodes Example 4 showstats Maui running for 22 01 00 stats initialized on Mon Mar 26 17 43 34 Eligible Idle Jobs 15 45 33333 Active Jobs 42 Successful Completed Jobs 873 875 99 7 Avg Max QTime Hours 2 71 4 50 Avg Max XFactor 1 03 4 79 Dedicated Total ProcHours 4353 55 4782 10 91 0383 Current Active Total Procs 183 192 95 312 Avg WallClock Accuracy 43 25 Avg Job Proc Efficiency 98 17 Est Avg Backlog Hours 34 5 41 8 This example shows a concise summary of the system scheduling state Note that showstats and showstats s are equivalent The first line of output indicates the number of sc
166. es will be associated with the account of the job using the resources When the resources are idle the resources will be charged to the reservation s charge account In the case of standing reservations this account is specified using the parameter SRCHARGEACCOUNT In the case of administrative reservations this account is specified via a command line flag to the setres command Maui will only interface to the allocations bank when running in NORMAL mode However this behavior can be overridden by setting the environment variable MAUIBANKTEST to any value With this variable set Maui will attempt to interface to the bank in both SIMULATION and TEST mode The allocation manager interface allows you to charge accounts in a number of different ways Some sites may wish to charge for all jobs run through a system regardless of whether or not the job completed successfully Sites may also want to charge based on differing usage metrics such as walltime dedicated or processors actually utilized Maui supports the following charge policies specified via the parameter BANKCHARGEPOLICY DEBITALLWC charge for all jobs regardless of job completion state using processor weighted wallclock time dedicated as the usage metric DEBITSUCCESSFULWC charge only for jobs which successfully complete using processor weighted wallclock time dedicated as the usage metric DEBITSUCCESSFULCPU charge only for jobs which successfully complete using CPU time
167. ete colon delimited list of hosts allocated to job i e APPEL Resource manager specific list of job attributes if specified See the Resource Manager Extension Overview for more info lt STRING gt lt STRING gt NONE node001 node004 NOTE In Maui 3 0 this field only lists the job s master host 39 lt STRING gt NONE Name of resource manager if specified List of hosts required by job if taskcount gt hosts scheduler must use these nodes in addition to others if taskcount lt host scheduler must select needed hosts from this list lt STRING gt NONE Name of reservation required by job if specified Set constraints required by node in the form lt SetConstraint gt lt SetType gt lt SetList gt where SetConstraint is one of ONEOF FIRSTOF or ANYOBF SetType is one of PROCSPEED FEATURE or NETWORK and SetList is an optional colon delimited list of allowed set attributes i e ONEOF PROCSPEED 350 450 500 Name of application simulator module and associated configuration data i e HSM IN infile txt 140000 OUT outfile txt 500000 lt STRING gt NONE RESERVED FOR FUTURE USE NOTE if no applicable value is specified the exact string NONE should be entered lt STRING gt lt STRING gt NONE lt STRING gt lt STRING gt lt STRING gt NONE lt STRING gt lt STRING gt
168. eueName Nodes MaxWallTime Priority ChargeRate Test lt 16 00 30 00 100 Lx Serial 1 2200200 10 1x Serial Long 1 24 00 00 10 2x SHOrE ANG 4 00 00 10 Lx Short Long 2 16 24 00 00 10 2x Med 17 64 8 00 00 20 1x Med Long 17 64 24 00 00 20 23 Large 65 80 24 00 00 50 2X LargeMem 1 8 00 00 10 4x For charging management has decided to charge by job walltime since the nodes will not be shared Management has also dictated that 16 of the uniprocessor nodes should be dedicated to running small jobs requiring 16 or fewer nodes Management has also decided that it would like to allow only serial jobs to run on the large memory nodes and would like to charge these jobs at a rate of 4x There are no constraints on the remaining nodes Goals Should do This site has goals which are focused more on a supplying a straightforward queue environment to the end users than on maximizing the scheduling performance of the system The Maui configuration has the primary purpose of faithfully reproducing the queue constraints above while maintaining reasonable scheduling performance in the process Analysis Since we are using PBS as the resource manager this should be a pretty straightforward process It will involve setting up an allocations manager to handle charging configuring queue priorities and creating a system reservation to manage the 16 processors dedicated to small jobs and another for managing the large memory nodes Configuration This site
169. f an account which is out of allocations and no fallback account has been specified RM Reject the resource manager refuses to start the job RM Failure the resource manager is experiencing failures Policy Violation the job violates certain throttling policies preventing it from running now and in the future No QOS Access the job does not have access to the QOS level it requests Jobs which are placed in a batch hold will show up within Maui in the state BatchHold Job Defer In most cases a job violating these policies will not be placed into a batch hold immediately Rather it will be deferred The parameter DEFERTIME indicates how long it will be deferred At this time it will be allowed back into the idle queue and again considered for scheduling If it again is unable to run at that time or at any time in the future it is again deferred for the timeframe specified by DEFERTIME A job will be released and deferred up to DEFERCOUNT times at which point the scheduler places a batch hold on the job and waits for a system administrator to determine the correct course of action Deferred jobs will have a Maui state of Deferred As with jobs in the BatchHold state the reason the job was deferred can be determined by use of the checkjob command At any time a job can be released from any hold or deferred state using the releasehold command The Maui logs should provide detailed information about the cause of any batch hold or job d
170. f it is available but if not run the job as soon as possible anyway Finally the NODESETLIST value of switchA switchB tells Maui to only use node sets based on the listed feature values This is necessary since sites will often use node features for many purposes and the resulting node sets would be of little use for switch proximity if they were generated based on irrelevant node features indicating things such as processor speed or node architecture On occasion sites may wish to allow a less strict interpretation of nodes sets In particular many sites seek to enforce a more liberal PROCSPEED based node set policy where almost balanced node allocations are allowed but wildly varying node allocations are not In such cases the parameter NODESETTOLERANCE may be used This parameter allows specification of the percentage difference between the fastest and slowest node which can be within a nodeset using the following calculation Speed Max Speed Min Speed Min lt NODESETTOLERANCE Thus setting NODESETTOLERANCE to 0 5 would allow the fastest node in a particular node set to be up to 50 faster than the slowest node in that set With a 0 5 setting a job may allocate a mix of 500 and 750 MHz nodes but not a mix of 500 and 900 MHz nodes Currently tolerances are only supported when the NODESETATTRIBUTE parameter is set to PROCSPEED The MAXBALANCE node allocation algorithm is often used in conjunction with tolerance based node sets
171. f the following approaches Direct resource manager specification Some node attribute may be directly specified through the resource manager For example Loadleveler allows a site to assign a MachineSpeed value to each node If the site chooses to specify this value within the Loadleveler configuration Maui will obtain this info via the Loadleveler scheduling API and use it in scheduling decisions The list of node attributes supported in this manner varies from resource manager to resource manager and should be determined by consulting resource manager documentation Translation of resource manager specified opaque attributes Many resource managers support the concept of opaque node attributes allowing a site to assign arbitrary strings to anode These strings are opaque in the sense that the resource manager passes them along to the scheduler without assigning any meaning to them Nodes possessing these opaque attributes can then be requested by various jobs Using certain Maui parameters sites can assign a meaning within Maui to these opaque node attributes and extract specific node information For example setting the parameter FEATUREPROCSPEEDHEADER xps will cause a node with the opaque string xps950 to be a assigned a processor speed of 950 MHz within Maui Default node attributes Some default node attributes can be assigned on a frame or partition basis Unless explicitly specified otherwise nodes within the parti
172. f the resource cluster supercomputer determine how they want the system to be used often via an allocations committee over a particular timeframe often a month quarter or year To enforce their decisions they distribute allocations to various projects via various accounts and assign each account an account manager These allocations can be for use particular machines or globally usable They can also have activation and expiration dates associated with them All transaction information is typically stored in a database or directory server allowing extensive statistical and allocation tracking Each account manager determines how the allocations are made available to individual users within his project Allocation manager managers such as PNNL s QBank allow the account manager to dedicate portions of the overall allocation to individual users specify some of allocations as shared by all users and hold some of the allocations in reserve for later use When using an allocations manager each job must be associated with an account To accomplish this with minimal user impact the allocation manager could be set up to handle default accounts on a per user basis However as is often the case some users may be active on more than one project and thus have access to more than one account In these situations a mechanism such as a job command file keyword should be provided to allow a user to specify which account should be associated with the job
173. f the timeframe covered by the displayed statistics The statistical output is divided into two categories Running and Completed Running statistics include information about jobs that are currently running Completed statistics are compiled using historical information from both running and completed jobs The fields are as follows GroupName Name of group GID Group ID of group Jobs Number of running jobs Procs Number of procs allocated to running jobs ProcHours Number of proc hours required to complete running jobs Jobs Number of jobs completed Percentage of total jobs that were completed by group PHReq Total proc hours requested by completed jobs Percentage of total proc hours requested by completed jobs that were requested by group PHDed Total proc hours dedicated to active and completed jobs The proc hours dedicated to a job are calculated by multiplying the number of allocated procs by the length of time the procs were allocated regardless of the job s CPU usage o Percentage of total proce hours dedicated that were dedicated by group FSTgt Fairshare target A group s fairshare target is specified in the fs cfg file This value should be compared to the group s node hour dedicated percentage to determine if the target is being met AvgXF Average expansion factor for jobs completed A job s XFactor expansion factor is calculated by the following formula QueuedTime RunTime WallClockLimit MaxXF H
174. for resources they will generally decrease the performance of the cluster delay the execution of these jobs and possibly cause one or more of the jobs to fail The scheduler is responsible for internally tracking and dedicating requested resources to a job thus preventing use of these resources by other jobs 1 1 2 Mission Policies When clusters or other HPC platforms are created they are typically created for one or more specific purposes These purposes or mission goals often define various rules about how the system should be used and who or what will be allowed to use it To be effective a scheduler must provide a suite of policies which allow a site to map site mission policies into scheduling behavior 1 1 3 Optimizations The compute power of a cluster is a limited resource and over time demand will inevitably exceed supply Intelligent scheduling decisions can significantly improve the effectiveness of the cluster resulting in more jobs being run and quicker job turnaround Subject to the constraints of the traffic control and mission policies it is the job of the scheduler to use whatever freedom is available to schedule jobs in such a manner so as to maximize cluster performance Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 1 2 Philosophy and Goals of the Maui Scheduler Managers desire maximum return on investment often meaning high system utilization and the ability to deli
175. future reservations the MINRESOURCE algorithm is used 5 2 2 2 FIRSTAVAILABLE Simple first come first server algorithm where nodes are allocated in the order they are presented by the resource manager This is a very simple and very fast algorithm 5 2 2 3 LASTAVAILABLE Algorithm which selects resources so as to minimize the amount of time after the job and before the the trailing reservation This algorithm is a best fit in time algorithm which minimizes the impact of reservation based node time fragmentation It is useful in systems where a large number of reservations job standing or administrative are in place 5 2 2 4 MACHINEPRIO This algorithm allows a site to specify the priority of various static and dynamic aspects of compute nodes and allocate them accordingly It is in essence a flexible version of the MINRESOURCE algorithm 5 2 2 5 MINRESOURCE This algorithm priorities nodes according to the configured resources on each node Those nodes with the fewest configured resources which still meet the job s resource constraints are selected 5 2 2 6 CONTIGUOUS This algorithm will allocate nodes in contiguous linear blocks as required by the Compaq RMS system 5 2 2 7 MAXBALANCE This algorithm will attempt to allocate the most balanced set of nodes possible to a job In most cases but not all the metric for balance of the nodes is node speed Thus if possible nodes with identical speeds will be allocated
176. g Cone a iS SIMJOBSUBMISSIONPOLICY NORMAL SIMJOBSUBMISSIONPOLICY AASE CONSTANTJOBDEPTH or CONSTANTJOBDEPTH ia d CONSTANTPSDEPTH Maui will submit jobs with the relative time CONSTANTPSDEPTH attempt to maintain an idle distribution specified in the workload trace file queue of lt SIMINITIALQUEUEDEPTH gt jobs and procseconds respectively specifies whether or not maui one of the following will filter nodes based on SIMNODECONFIGURATION UNIFORM or NORMAL NORMAL resource configuration while running a simulation specifies the maximum number SIMNODECOUNT lt INTEGER gt 0 no limit of nodes maui will load from the simulation resource file specifies the file from which maui will obtain node SIMRESOURCETRACEFILE traces nodes 1 information when running in oe wae SIMRESOURCETRACEFILE lt STRING gt traces resource trace simulation mode Maui will Maui will obtain node traces when running in attempt to locate the file relative Simulation mode from the to lt MAUIHOMEDIR gt unless lt MAUIHOMEDIR gt traces nodes 1 file specified as an absolute path specifies the random delay p added to the RM command base SIMRMRANDOMDELAY 5 SIMRMRANDOMDELAY lt INTEGER gt 0 delay accumulated when making Maui will add a random delay of between 0 and 5 any resource manager call in seconds to the simulated time delay of all RM calls simulation mode specifies on which scheduling SIMSTOPITERATION 1 iteration a maui simulation will SIMSTOPITERATION lt IN
177. g targets are met The priority calculation for the target factor is Priority TARGWEIGHT QueueTimeComponent XFactorComponent The queue time and expansion factor target are specified on a per QOS basis using the QOSXFTARGET and QOSQTTARGET parameters The QueueTime and XFactor component calculations are designed produce small values until the target value begins to approach at which point these components grow very rapidly If the target is missed these component will remain high and continue to grow but will not grow exponentially 5 1 2 6 Usage USAGE Component Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 5 1 3 Common Priority Usage Sites vary wildly in the preferred manner of prioritizing jobs Maui s scheduling hierarchy allows sites to meet their job control needs without requiring them to adjust dozens of parameters Some sites may choose to utilize numerous subcomponents others a few and still others are completely happy with the default FIFO behavior Any subcomponent which is not of interest may be safely ignored To help clarify the use of priority weights a brief example may help Suppose a site wished to maintain the FIFO behavior but also incorporate some credential based prioritization to favor a special user Particularly the site would like the userjohn to receive a higher initial priority than all other users Configuring this beh
178. g the Message Buffer 14 4 Handling Events with the Notification Routine 14 5 Issues with Client Commands 14 6 Tracking System Failures 14 7 Problems with Individual Jobs 15 0 Improving User Effectiveness 15 1 User Feedback Loops 15 2 User Level Statistics 15 3_ Enhancing Wallclock Limit Estimates 15 4 Providing Resource Availability Information 15 5 Job Start Time Estimates 15 6 Collecting Performance Information on Individual Jobs 16 0 Simulations 16 1 Simulation Overview 16 2 Resource Traces 16 3 Workload Traces 16 4 Simulation Specific Configuration 17 0 Miscellaneous 17 1 User Feedback Appendices Appendix A Case Studies Appendix B Extension Interface Appendix C Adding New Algorithms Appendix D Adjusting Default Limits Appendix E Security Configuration Appendix F Parameters Overview Appendix G Commands Overview Acknowledgements Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 1 0 Philosophy The goal of a scheduler in the broadest sense is to make users administrators and managers happy Users desire the ability to specify resources obtain quick turnaround on their jobs and receive reliable allocation of resources Administrators desire happy managers and happy users They also desire the ability to understand both the workload and the resources available This includes current state problems and statistics as well as information about what is happening under the
179. g this is Fairshare This feature can be used to adjust the priority of jobs to favor disfavor jobs based on fairshare targets and historical usage In essence this feature improves the turnaround time of the jobs not meeting their fairshare target at the expense of those that are Depending on the criticality of the delivered cycle distribution constraints this site might also wish to consider an allocations bank such as PNNL s QBank which enables more stringent control over the amount of resources which can be delivered to various users To manage the primetime job turnaround constraints a standing reservation would probably be the best approach A standing reservation can be used to set aside a subset of the nodes for quick turnaround jobs This reservation can be configured with a time based access point to allow only jobs which will complete within some time X to utilize these resources The reservation has advantages over a typical queue based solution in this case in that these quick turnaround jobs can be run anywhere resources are available either inside or outside the reservation or even crossing reservation boundaries The site does not have any hard constraints about what is acceptable turnaround time so the best approach would probably be to analyze the site s workload under a number of configurations using the simulator and observe the corresponding scheduling behavior For general optimization there are a number of scheduling aspects
180. ging information Logging information will be written in the file lt MAUIHOMEDIR gt lt LOGDIR gt lt LOGFILE gt unless lt LOGDIR gt or lt LOGFILE gt is specified using an absolute path If the log file is not specified or points to an invalid file all logging information is directed to STDERR However because of the sheer volume of information that can be logged it is not recommended that this be done while in production By default LOGDIR and LOGFILE are set to log and maui log respectively resulting in scheduler logs being written to lt MAUIHOMEDIR gt log maui log The parameter LOGFILEMAXSIZE determines how large the log file is allowed to become before it is rolled and is set to 10 MB by default When the log file reaches this specified size the log file is rolled The parameter LOGFILEROLLDEPTH will control the number of old logs maintained and defaults to 1 Rolled log files will have a numeric suffix appended indicating their order The parameter LOGLEVEL controls the verbosity of the information Currently LOGLEVEL values between 0 and 9 are used to control the amount of information logged with 0 being the most terse logging only the most server problems detected while 9 is the most verbose commenting on just about everything The amount of information provided at each log level is approximately an order of magnitude greater than what is provided at the log level immediately below it A LOGLEVEL of 2 will record virtu
181. gt isone A Maui NOTE for RMTYPE PYMOST UT cluster ERMS WIKI RMAUTHTYPE must RMPORT 1 15004 be set to CHECKSUM Maui will interface to two different PBS resource managers one located on server cluster1 at port 15003 and one located on server cluster2 at port 15004 hostname of machine on which SERVERHOST geronimo scc edu SERVERHOST lt HOSTNAME gt NONE maui will run NOTE this Maui will execute on the host parameter MUST be specified geronimo scc edu one of the following specifies how Maui interacts SERVERMODE NORMAL with the outside world See SERVERMODE SIMULATION NORMAT TEST cor SIMULATION lt Testing gt for more information specifies the name the scheduler SERVERNAME lt STRING gt lt SERVERHOST gt alise toreter to tsel at communication with peer daemons SERVERPORT 30003 port on which maui will open its SERVERPORT lt INTEGER gt range 1 64000 40559 d r interface soket Maui will listen for client socket connections on port 30003 if TRUE the scheduler will end SIMAUTOSHUTDOWN ON simulations when the active SIMAUTOSHUTDOWN lt BOOLEAN gt TRUE queue and idle queue become The scheduler simulation will end as soon as there are empty no jobs running and no idle jobs which could run specifies whether to increase or SIMCPUSCALINGPERCENT lt INTEGER gt 100 no scaling Cee runtime and wallclock limit of each job in the workload trace file zero or more of the following cause Maui to force the SIMDEFAULTJOBFLAGS D
182. gt more scheduling iterations have been completed Example Shut maui down gt schedctl k maui shutdown Example Stop maui scheduling gt schedctl s maui will stop scheduling immediately Example Resume maui scheduling gt schedctl r maui will resume scheduling immediately Example Stop maui scheduling in 100 more iterations Specify that maui should not respond to client requests until that point is reached gt schedctl S 100I maui will stop scheduling in 100 iterations Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ sethold sethold b h JOB JOB JOB Purpose Set hold on specified job s Permissions This command can be run by any Maui Scheduler Administrator Parameters JOB Job number of job to hold Flags b Set a batch hold Typically only the scheduler places batch holds This flag allows an administrator to manually set a batch hold h Help for this command Description This command allows you to place a hold upon specified jobs Example sethold b fr1l7n02 1072 0 fr1i5n03 1017 0 Batch Hold Placed on All Specified Jobs In this example a batch hold was placed on job fr17n02 1072 0 and job fr15n03 1017 0 Related Commands Release holds with the releasehold command Default File Location u loadl maui bin sethold Notes None Copyright 1998 Maui High Performance Computing Center All rights reserved Cop
183. guration when Maui decides to start a job it contacts the bank and requests an allocation reservation or lien be placed on the associated account This allocation reservation is equivalent to the total amount of allocation which could be consumed by the job based on the job s wallclock limit and is used to prevent the possibility of allocation oversubscription Maui then starts the job When the job completes Maui debits the amount of allocation actually consumed by the job from the job s account and then releases the allocation reservation or lien These steps transpire under the covers and should be undetectable by outside users Only when an account has insufficient allocations to run a requested job will the presence of the allocation bank be noticed If desired an account may be specified which is to be used when a job s primary account is out of allocations This account specified using the parameter BANKFALLBACKACCOUNT is often associated with a low QOS privilege set and priority and often is configured to only run when no other jobs are present Reservations can also be configured to be chargeable One of the big hesitations have with dedicating resources to a particular group is that if the resources are not used by that group they go idle and are wasted By configuration a reservation to be chargeable sites can charge every idle cycle of the reservation to a particular project When the reservation is in use the consumed resourc
184. h index values can be numeric or alphanumeric strings If no array index is specified for an array parameter an index of 0 is assumed See the parameters documentation for information on specific parameters All config files are read when Maui is started up Also the schedctl R command can be used to reconfigure the scheduler at any time forcing it to re read all config files before continuing The command changeparam can be used to change individual parameter settings at any time 1 e changeparam LOGLEVEL 3 Changes made by the changeparam command are not persistent so will be overwritten the next time the config file values are loaded The current parameter settings can be viewed at any time using the showconfig command Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved gt Maui Commands 4 1 Client Overview 4 2 Monitoring System Status 4 3 Managing Jobs 4 4 Managing Reservations 4 5 Configuring Policies 4 6 End User Commands CECECECECA 4 1 Miscellaneous Commands Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 4 1 Client Overview Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 4 2 Status Commands Maui provides an array of commands to organize and present information about the current state and historical statistics of the scheduler jobs resources users accounts etc Th
185. h and Development Group All Rights Reserved 7 1 5 Configuring and Managing Reservations All reservations whether they be administrative or standing possess many similar traits 7 1 5 1 Reservation Attributes All reservations possess a timeframe of activity an access control list and a list of resources to be reserved Additionally reservations may also possess a number of extension flags which modify the behavior of the reservation 7 1 5 1 1 Start End Time All reservations possess a start and an end time which define the reservation s active time During this active time the resources within the reservation may only be used as specified by the reservation ACL This active time may be specified as either a start end pair or a start duration pair Reservations exist and are visible from the time they are created until the active time ends at which point they are automatically removed 7 1 5 1 2 Access Control List ACL For a reservation to be useful it must be able to limit who or what can access the resources it has reserved This is handled by way of an access control list or ACL 7 1 5 1 3 Resources When specifying which resources to reserve the administrator has a number of options These options allow control over how many resources are reserved and where they are reserved at The following reservation attributes allow the administrator to define resources Task Description A key concept of reservations is the
186. heduling iterations performed by the current scheduling process followed by the time the scheduler started The second line indicates the amount of time the Maui Scheduler has been scheduling in HH MM SS notation followed by the statistics initialization time The fields are as follows Active Jobs Number of jobs currently active Running or Starting Eligible Jobs Number of jobs in the system queue jobs that are considered when scheduling Idle Jobs Number of jobs both in and out of the system queue that are in the LoadLeveler Idle state Completed Jobs Number of jobs completed since statistics were initialized Successful Jobs Jobs that completed successfully without abnormal termination XFactor Average expansion factor of all completed jobs Max XFactor Maximum expansion factor of completed jobs Max Bypass Maximum bypass of completed jobs Available ProcHours Total proc hours available to the scheduler Dedicated ProcHours Total proc hours made available to jobs Effic Scheduling efficiency DedicatedProcHours Available ProcHours Min Efficiency Minimum scheduling efficiency obtained since scheduler was started Iteration Iteration on which the minimum scheduling efficiency occurred Available Procs Number of procs currently available Busy Procs Number of procs currently busy Effic Current system efficiency BusyProcs AvailableProcs WallClock Accuracy Average wall clock accuracy of completed jobs job weighted average Job
187. hernet Features Thin Dedicated Classes batch medium Frame 26 Node 10 StateTime Node has been in current state for 5 02 23 DownTime 26844 Seconds 7 46 Hours Thu Sep 4 09 00 00 Load 0 009 TotalTime 30 18 29 UpTime 23 28 51 77 47 BusyTime 19 21 46 63 89 Related Commands Further information about node status can be found using the showst ate command You can determine scheduling nodes with the LoadLeveler 11st atus command nodes that have Avail in the Schedd column Default File Location u loadl maui bin checknode Notes None Copyright 1998 Maui High Performance Computing Center All rights reserved diagnose Under Construction Overview The diagnose command is used to display information about various aspects of scheduling and the results of internal diagnostic tests Format diagnose a lt ACCOUNTID gt Diagnose Tar Diagnose g lt GROUPID gt Diagnose I J XJOBID gt 1 Diagnose m Diagnose f t lt PARTITION gt lt NODEID gt Nodes I p t lt PARTITION gt Diagnose q 1 lt POLICYLEVEL gt Diagnose Taul Diagnose Configuration Diagnose Reservations Tae Diagnose Partitions I u lt USERID gt Diagnose Flags a Show detailed information about accounts f Show detailed information about fairshare configuration and status j Show detailed information
188. holes in node space backfill tends to favor smaller and shorter running jobs more than larger and longer running ones It is common to see over 90 of these small and short jobs backfilled Consequently sites will see marked improvement in the level of service delivered to the small short jobs and only moderate to no improvement for the larger long ones The question arises is backfill a purely good feature Doesn t there have to be a trade off some where Doesn t there have to be a dark side Well there are a few drawbacks to using backfill but they are fairly minor First of all because backfill locates jobs to run scattered throughout the idle job queue it tends to diminish the influence of the job prioritization a site has chosen and thus may negate any desired workload steering attempts through this prioritization Secondly although the start time of the highest priority job is protected by a reservation what is to prevent the third priority job from starting early and possibly delaying the start of the second priority job Ahh a problem Actually one that is easily handled as will be described later The third problem is actually a little more subtle Consider the following scenario involving the 2 processor cluster shown in figure 1 Job A has a 4 hour wallclock limit and requires 1 processor It started 1 hour ago and will reach its wallclock limit in 3 more hours Job B is the highest priority idle job and requires 2 processors for
189. hould be compared to the user s node hour dedicated percentage to determine if the target is being met AvgXF Average expansion factor for jobs completed A job s XFactor expansion factor is calculated by the following formula QueuedTime RunTime WallClockLimit MaxXF Highest expansion factor received by jobs completed AvgQH Average queue time in hours of jobs Effic Average job efficiency Job efficiency is calculated by dividing the actual node hours of CPU time used by the job by the node hours allocated to the job WCAcc Average wall clock accuracy for jobs completed Wall clock accuracy is calculated by dividing a job s actual run time by its specified wall clock limit These fields are empty until a user has completed at least one job Related Commands Use the reset stats command to re initialize statistics Notes See the Statistics document for more details about scheduler statistics Copyright 1998 Maui High Performance Computing Center All rights reserved Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved
190. how the settings of all parameters It does show all major parameters and all parameters which are in effect and have been set to non default values However it hides other rarely used parameters and those which currently have no effect or are set to default values To show the settings of all parameters use the v verbose flag This will provide an extended output This output is often best used in conjunction with the grep command as the output can be voluminous Related Commands Use the changeparam command to change the various Maui Scheduler parameters Notes See the Parameters document for details about configurable parameters Copyright 1998 Maui High Performance Computing Center All rights reserved Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ showgrid showgrid STATISTICTYPE h Purpose Shows table of various scheduler statistics Permissions This command can be run by any Maui Scheduler Administrator Parameters STATISTICTYPE Values for this parameter AVGBYPASS Average bypass count Includes summary of job weighted expansion bypass and total samples AVGQTIME Average queue time Includes summary of job weighted queue time and total samples AVGXFACTOR Average expansion factor Includes summary of job weighted expansion factor node weighted expansion factor node second weighted expansion factor and total number of samples BFCOUNT Number of jobs ba
191. ibutes such as job owner job size length of time the job has been queued and so forth 3 3 2 3 Enforce Configured Throttling Policies Any configured throttling policies are then applied constraining how many jobs nodes processors etc are allowed on a per credential basis Jobs which violate these policies are not considered for scheduling 3 3 2 4 Determine Resource Availability For each job Maui attempts to locate the required compute resources needed by the job In order for a match to be made the node must possess all node attributes specified by the job and possess adequate available resources to meet the TasksPerNode job constraint Default TasksPerNode is 1 Normally Maui determine a node to have adequate resources if the resources are neither utilized by nor dedicated to another job using the calculation R Available R Configured MAX R Dedicated R Utilized The RESOURCEAVAILABILITYPOLICY parameter can be modified to adjust this behavior 3 3 2 5 Allocate Resources to Job If adequate resources can be found for a job the node allocation policy is then applied to select the best set of resources These allocation policies allow selection criteria such as speed of node type of reservations or excess node resources to be figured into the allocation decision to improve the performance of the job and or maximize the freedom of the scheduler in making future scheduling decisions 3 3 2 6 Distribute Jobs Tasks Acro
192. ied by the BACKFILLPOLICY parameter be it FIRSTFIT BESTFIT Time gt etc Backfillable Nodes Nodes Assuming the BESTFIT algorithm is applied the following steps are taken 1 The list of feasible Window 1 backfill jobs is filtered selecting only those which will actually fit in 3 hs te cael window 2 The degree of fit of each job is determined based on the SCHEDULINGCRITERIA parameter ie processors seconds processor seconds etc ie if processors is selected the job which requests the most processors will have the best fit 3 The job with the best fit is started 4 While backfill jobs and idle resources remain repeat step 1 Other backfill policies behave in a generally similar manner The parameters documentation can provide further details One final important note By default Maui reserves only the highest priority job resulting in a very liberal and aggressive backfill This reservation guarantees that backfilled jobs will not delay the highest priority job although they may delay the second highest priority job Actually due to wallclock inaccuracies it is possible the the highest priority job may actually get slightly delayed as well but we won t go into that The parameter RESERVATIONDEPTH controls how conservative liberal the backfill policy is This parameter controls how deep down the priority queue to make reservations While increasing this parameter wil
193. ies which users have SRUSERLIST 1 bob joe mary SRUSERLIST X space delimited list of users NONE access to the resources reserved users bob joe and mary can all access the resources by this reservation reserved within this reservation SRSTARTTIME 1 1 08 00 00 specifies the week offset at SRENDTIME 1 5 17 00 00 SRWENDTIME X DD HH MM SS 7 00 00 00 which the stand reservation should end standing reservation 1 will run from Monday 8 00 AM to Friday 5 00 PM SRSTARTTIME 1 1 08 00 00 specifies the week offset at SRENDTIME 1 5 17 00 00 SRWSTARTTIME X DD HH MM SS 0 00 00 00 which the standing reservation should start standing reservation 1 will run from Monday 8 00 AM to Friday 5 00 PM specifies the directory in which STATDIR lt STRING gt stats Maui statistics will be STATDIR var adm maui stats maintained list of zero or more space delimited specifies system wide default lt ATTR gt lt VALUE gt pairs where lt ATTR gt attributes See the SYSCFG PLIST Partitionl QDEF highprio is one of the following Attribute Flag Overview for ve SYSCPG PRIORITY FSTARGET QLIST QDEF NONE more information A GS o ae be aa ao ae PLIST PDEF FLAGS or a fairness policy NOTE Only available in Maui specification 3 0 7 and higher specifies the priority weight SWAPWEIGHT lt INTEGER gt 0 assigned to the virtual memory SWAPWEIGHT 10 request of a job specifies the walltime for jobs SYSTEMDEFAULTJOBWALLTIME 1 00 00 00 SYSTEMDEFAULT
194. ighest expansion factor received by jobs completed AvgQH Average queue time in hours of jobs Effic Average job efficiency Job efficiency is calculated by dividing the actual node hours of CPU time used by the job by the node hours allocated to the job WCAcc Average wall clock accuracy for jobs completed Wall clock accuracy is calculated by dividing a job s actual run time by its specified wall clock limit These fields are empty until a group has completed at least one job Example 3 o showstats n S Memory Requirement Breakdown Memory Nodes Percent InitialNH Percent NodeHours Percent 64 8 2 78 9799 794 92 1232 100 00 128 144 50 00 9162 41 29 22190 100 00 256 32 HL 3 ab 20290 411 47 4931 100 00 512 96 33 33 5080 34 34 14793 100 00 1024 8 22 13 48 3 89 1232 100 00 2048 0 0 00 0 0 00 0 0 00 TOTAL 288 100 00 44381 100 00 44381 100 00 Node Statistics Summary 8 64MB Nodes 99 26 Avail 79 18 Busy Current 100 00 Avail 100 00 Busy Summary 144 128MB Nodes 98 99 Avail 75 92 Busy Current 100 00 Avail 100 00 Busy Summary 32 256MB Nodes 97 69 Avail 85 66 Busy Current 100 00 Avail 100 00 Busy Summary 96 512MB Nodes 96 12 Avail 82 92 Busy Current 98 96 Avail 94 79 Busy Summary 8 1024MB Nodes 99 87 Avail 81 77 Busy Current 100 00 Avail 75 00 Busy System Summary 288 Nodes 97 92 Avail 79 59 Busy Current 99 65 Avail 97 57 Busy This example shows a statistical listing of nod
195. imum allowed total PE count which can be dedicated at any given time See note for Maui 3 0 versions maximum allowed total PE count which can be dedicated at any given time See note for Maui 3 0 versions maximum allowed total processors which can be dedicated at any give time See note for Maui 3 0 versions maximum allowed sum of outstanding dedicated processor second obligations of all active jobs See note for Maui 3 0 versions maximum allowed sum of outstanding walltime limits of all active jobs NOTE only available in Maui 3 2 and higher specifies the coefficient to be multiplied by a job s MEM dedicated memory in MB factor LOGFILEROLLDEPTH 5 Maui will maintain the last 5 log files LOGLEVEL 4 Maui will write all Maui log messages with a threshold of 4 or lower to the maui log file RESWEIGH 10 MEMWEIGHT 0 1000 each job s priority will be increased by 10 1000 its MEM factor one of the following DEDICATED specifies whether or not Maui will allow node resources to be NODEACCESSPOLICY SHARED NODEACCESSPOLICY SHARED or SINGLEUSER DEDICATED shared or dedicated by Maui will allow resources on a node to be used by independent jobs more than one job one of the following FIRSTAVAILABLE s
196. ion Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 13 3 Resource Manager Extensions All resource managers are not created equal There is a wide range in what capabilities are available from system to system Additionally there is a large body of functionality which many if not all resource managers have no concept of A good example of this is job QoS Since most resource managers do not have a concept of quality of service they do not provide a mechanism for users to specify this information In many cases Maui is able to add capabilities at a global level However a number of features require a per job specification Resource manager extensions allow this information to be associated with the job How this is done varies with the resource manager Both Loadleveler and Wiki allow the specification of a comment field In Loadleveler specified as comment lt X gt gt PBS does not support this ability by default but is extendible via the W flag see the PBS Resource Manager Extension Overview Using the resource manager specific method the following job extensions are currently available Default sogi Name Format Value Description Example ACCESSMODE one of DEDICATED or SHARED SHARED ACCESSMODE DEDICATED dedicated DMEM lt INTEGER gt 0 memory per DMEM 512 task in MB one or more of the following comma separated keywords FLAGS ADVRES RESID RESTARTABLE
197. ite reason If the logging level is configured too high huge volumes of log output may be recorded potentially obscuring the problems in a flood of data Intelligent searching combined with the use of the LOGLEVEL and LOGFACILITY parameters can mine out the needed information Key information associated with various problems is generally marked with the keywords WARNING ALERT or ERROR See the Logging Overview for further information Using a Debugger If other methods do not resolve the problem the use of a debugger can provide missing information While output recorded in the Maui logs can specify which routine is failing the debugger can actually locate the very source of the problem Log information can help you pinpoint exactly which section of code needs to be examined and which data is suspicious Historically combining log information with debugger flexibility have made locating and correcting Maui bugs a relatively quick and straightforward process To use a debugger you can either attach to a running Maui process or start Maui under the debugger Starting Maui under a debugger requires that the MAUIDEBUG environment variable be set to the value yes to prevent Maui from daemonizing and backgrounding itself The following example shows a typical debugging start up using gdb gt export MAUIDEBUG yes gt cd lt MAUIHOMEDIR gt src gt gdb bin maui gt b QOSInitialize gt The gdb debugger has the ability to specify c
198. job the job could utilize these resources according to its own task description and needs Currently the resources which can be associated with reservations include processors memory swap local disk initiator classes and any number of arbitrary resources Arbitrary resources may include peripherals such as tape drives software licenses or any other site specific resource 7 1 1 2 TimeFrame Associated with each reservation is a timeframe This specifies when the resources will be reserved or dedicated to jobs which meet the reservation s ACL The timeframe simply consists of a start time and an end time When configuring a reservation this information may be specified as a start time together with either an end time or a duration 7 1 1 3 Access Control List A reservation s access control list specifies which jobs can use a reservation Only jobs which meet one or more of a reservation s access criteria are allowed to use the reserved resources during the reservation timeframe Currently the reservation access criteria include the following users groups accounts classes QOS and job duration 7 1 1 4 Job to Reservation Mapping While a reservation s ACL will allow particular jobs to utilize reserved resources it does not force any job to utilize these resources With each job Maui attempts to locate the best possible combination of available resources whether these are reserved or unreserved For example in the figure bel
199. l improve guarantees that priority jobs will not be bypassed it reduces the freedom of the scheduler to backfill resulting in somewhat lower system utilization The value of the trade offs often need to be determined on a site by site basis 8 2 3 Configuring Backfill Backfill is enabled in Maui by specifying the BACKFILLPOLICY parameter By default backfill is enabled in Maui using the FIRSTFIT algorithm However this parameter can also be set to BESTFIT GREEDY or NONE The number of reservations can also be controlled using RESERVATIONDEPTH lt X gt This depth can be distributed across job QOS levels using RESERVATIONQOSLIST lt X gt See also Parameters BACKFILLDEPTHand BACKFILLMETRIC Reservation Policy Overview Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 8 3 Node Set Overview While backfill improves the scheduler s performance this is only half the battle The efficiency of a cluster in terms of actual work accomplished is a function of both scheduling performance and individual job efficiency In many clusters job efficiency can vary from node to node as well as with the node mix allocated Most parallel jobs written in popular languages such as MPI or PVM do not internally load balance their workload and thus run only as fast as the slowest node allocated Consequently these jobs run most effectively on homogeneous sets of nodes However while many clusters start out
200. lculations specifies the weight to be applied to a job s minimum XFWEIGHT lt INTEGER gt 0 expansion factor before it is added to the job s cumulative priority XFWEIGHT 1000 Maui will multiply a job s XFactor value by 1000 and then add this value to its total priority Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved Appendix G Commands Overview Command Description checkjob provide detailed status report for specified job checknode provide detailed status report for specified node Aide provide diagnostic report for various aspects of resources workload and a scheduling Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved G 1 canceljob canceljob JOB JOB h Purpose Cancels the specified job s Permissions This command can be run by any Maui Scheduler Administrator and the owner of the job Parameters JOB Job name you want to cancel Flags h Show help for this command Description The cancel job command is used to selectively cancel the specified job s active idle or non queued from the queue Example 1 canceljob h Shows help for this command Example 2 canceljob frin04 981 0 Cancels job 981 running on Frame 1 Node 04 Related Commands This command is equivalent to the LoadLeveler 11cance1l command You can find job numbers with the showg
201. le and or active Maui fairness policies 1 e max job per user max idle job per user etc As a final note the parameter QUEUETIMEWEIGHT can be adjusted on a per QOS basis using the QOSCFG parameter and the QTWEIGHT attribute For example the line QOSCFG special QTWEIGHT 5000 will cause jobs utilizing the QOS special to have their queue time subcomponent weight increased by 5000 5 1 2 4 2 Expansion Factor XFACTOR Subcomponent The expansion factor subcomponent has an effect similar to the queue time factor but favors shorter jobs based on their requested wallclock run time In its canonical form the expansion factor XFactor metric is calculated as XFACTOR 1 lt QUEUETIME gt lt EXECUTIONTIME gt However a couple of aspects of this calculation make its use more difficult First the length of time the job will actually run Execution Time is not actually known until the job completes All that is known is how much time the job requests Secondly as described in the Queue Time Subcomponent section Maui does not necessarily use the raw time since job submission to determine QueueTime so as to prevent various scheduler abuses Consequently Maui uses the following modified equation XFACTOR 1 lt EFFQUEUETIME gt lt WALLCLOCKLIMIT gt In the equation above EF FQUEUETIME is the effective queue time subject to the JOBPRIOACCRUALPOLICY parameter and WALLCLOCKLIMIT is the user or system specified job wall
202. led state information and statistics for nodes that run jobs those running LoadL_startd NOTE This command returns an error message if it is run against a scheduling node one running schedd The following information is returned by this command Disk Disk space available Memory Memory available Swap Swap space available State Node state Opsys Operating system Arch Architecture Adapters Network adapters available Features Features available Classes Classes available Frame IBM SP frame number associated with node Node IBM SP node number associated with node StateTime Time node has been in current state in HH MM SS notation Downtime Displayed only if downtime is scheduled Load CPU Load Berkley one minute load average TotalTime Total time node has been detected since statistics initialization expressed in HH MM SS notation UpTime Total time node has been in an available Non Down state since statistics initialization expressed in HH MM SS notation percent of time up UpTime TotalTime BusyTime Total time node has been busy allocated to active jobs since statistics initialization expressed in HH MM SS notation percent of time busy BusyTime TotalTime After displaying this information some analysis is performed and any unusual conditions are reported Example checknode fr26n10 Checking Node fr26n10 mhpcc edu Disk KB 2076 Memory MB 512 Swap KB 470772 State Down Opsys AIX41 Arch R6000 Adapters et
203. licies lt TASKS Number of tasks actually allocated to job NOTE in lt INTEGER gt REQUESTED gt most cases this field is identical to field 3 Tasks Requested Tasks Per 23 lt INTEGER gt 1 Number of Tasks Per Node required by job or 1 if Node no requirement specified QOS requested delivered using the format QOS lt STRING gt lt STRING gt NONE lt QOS_REQUESTED gt lt QOS_DELIVERED gt ie hipriority bottomfeeder f square bracket delimited list of job attributes i e JobFlags 25 lt STRING gt lt STRING gt NONE BACKFILL BENCHMARK PREEMPTEE ne 26 lt STRING gt NONE Name of account associated with job if specified Executable 27 lt STRING gt NONE Name of job executable if specified Comment 28 lt STRING gt Bypass Number of time job was bypassed by lower priority Count Gale niece k jobs via backfill or 1 if not specified ca 30 lt DOUBLE gt 0 Number of processor seconds actually utilized by job Partition Name lt STRING gt DEFAULT Name of partition in which job ran 32 lt INTEGER gt Number of processors required per task lt INTEGER gt 0 Amount of RAM in MB required per task a lt INTEGER gt 0 Amount of local disk in MB required per task lt INTEGER gt 0 Amount of virtual memory in MB required per task Start Date 36 lt INTEGER gt 0 Epoch time indicating earliest time job can start End Date 37 lt INTEGER gt 0 Epoch time indicating latest time by which job must compl
204. lowed the scheduler to concurrently issue multiple node queries resulting in much quicker aggregate RM query times Responsiveness Finally in the non threaded serial approach the user interface was blocked while the scheduler updated various aspects of its workload resource and queue state In a threaded model the scheduler could continue to respond to queries and other commands even while fresh resource manager state information was being loaded resulting in much shorter average response times for user commands Under the threaded interface all resource manager information is loaded and processed while the user interface is still active Average aggregate resource manager API query times are tracked and new RM updates are launched so that the RM query will complete before the next scheduling iteration should start Where needed the loading process uses a pool of worker threads to issue large numbers of node specific information queries concurrently to accelerate this process The master thread continues to respond to user commands until all needed resource manager information is loaded and either a scheduling relevant event has occurred or the scheduling iteration time has arrived At this point the updated information is integrated into Maui s state information and scheduling is performed 13 1 2 Resource Manager Specific Details Limitations Special Features Under Construction LL LL2 PBS Wiki Synchronizing Conflicting Info
205. lt Algorithms to handle these type of issues are currently available in the G2 extension library reservation based systems A reservation based system adds the time dimension into the node allocation decision With reservations node resources must be viewed in a type of two dimension node time space Allocating nodes to jobs fragments this node time space and makes it more difficult to schedule jobs in the remaining more constrained node time slots Allocation decisions should be made in such a way as top minimize this fragmentation and maximize the schedulers ability to continue to start jobs in existing slots See the figure to hopefully remove a small amount of the incoherency contained in the above sentences In this figure Job A and job B are already running A reservation X has been created some time in the future Assume that job A is 2 hours long and job B is 3 hours long Again two new single processor jobs are submitted C and D job C requires 3 hours of compute time while job D requires 5 hours Either job will just fit in the free space located above Job A or in the free space located below job B If job C is placed above Job A job D requiring 5 hours of time will be prevented from running by the presence of reservation X However if job C is placed below job B job D can still start immediately above Job A Hopefully this canned example demonstrates the importance of time based reservation information in making node alloca
206. lt GROUP_LIST gt h USAGE HELP n lt NAME gt p lt PARTITION gt q lt QUEUE_LIST gt ie CLASS_LIST Q lt QOSLIST gt r lt RESOURCE_DESCRIPTION gt s lt STARTTIME gt u lt USER_LIST gt x lt FLAGS gt NOTE only available in Maui 3 2 and higher Purpose Reserve resources for use by jobs with particular credentials or attributes Access This command can be run by level 1 and level 2 Maui administrators Parameters Name O Format O O Defat Description list of accounts that will be ACCOUNT _LIST lt STRING gt lt STRING gt NONE allowed access to the reserved resources specifies which credentials will be accountable for unused resources dedicated to the reservation list of classes that will be allowed access to the reserved resource CHARGE _ SPEC lt ACCOUNTS gt lt GROUP gt lt USER gt DD HH MM SS HH MM SS L MO DD YY ENDTIME or INFINITY DD HH MM SS which must FEATURE_LIST lt STRING gt lt STRING gt NONE be possessed by the reserved resources list of FLAGS lt STRING gt lt STRING gt NONE CLASS_LIST lt STRING gt lt STRING gt DURATION a absolute or relative time reservation will end not reservation flags See Managing Reservations for details MIRE GROUP_LIST lt STRING gt lt S
207. may allow a site to better meet site mission objectives improve fairness or even improve overall system utilization Resource based prioritization is valuable when you want to favor jobs based on the resources requested This is good in three main scenarios first when you need to favor large resource jobs because its part of your site s mission statement second when you want to level the response time distribution across large and small jobs small jobs are more easily backfilled and thus generally have better turnaround time and finally when you want to improve system utilization What Yes system utilization actually increases as large resource jobs are pushed to the front of the queue This keeps the smaller jobs in the back where they can be selected for backfill and thus increase overall system utilization Its a lot like the story about filling a cup with golf balls and sand If you put the sand in first it gets in the way when you try to put in the golf balls However if you put in the golf balls first the sand can easily be poured in around them completely filling the cup The calculation for determining the total resource priority factor is Priority RESWEIGHT MIN RESOURCECAP NODEWE TIGHT TotalNodesRequested PROCWEIGHT TotalProcessorsRequested MEMWE TIGHT TotalMemoryRequested SWAPWE TIGHT TotalSwapRequested DISKWEIGHT TotalDiskRequested PEWEIGHT TotalPEHRequested The
208. ment for more information Copyright 1998 Maui High Performance Computing Center All rights reserved Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved showcontig showconfig v h Purpose View the current configurable parameters of the Maui Scheduler Permissions This command can be run by a level 1 2 or 3 Maui administrator Parameters None Flags h Help for this command v This optional flag turns on verbose mode which shows all possible Maui Scheduler parameters and their current settings If this flag is not used this command operates in context sensitive terse mode which shows only relevant parameter settings Description The showconfig command shows the current scheduler version and the settings of all in memory parameters These parameters are set via internal defaults command line arguments environment variable settings parameters in the maui cfg file and via the changeparam command Because of the many sources of configuration settings the output may differ from the contents of the maui cfg file The output is such that it can be saved and used as the contents of the maui cfg file if desired Example gt showconfig maui Scheduler version 3 0 2 0 PID 11080 BACKFILLPOLICY FIRSTFIT BACKFILLMETRIC NODES ALLOCATIONPOLICY MINRESOURCE RESERVATIONPOLICY CURRENTHIGHEST IMPORTANT NOTE the showconfig flag without the v flag does not s
209. minimum amount of time required to run this workload using the currently available nodes on the system Related Commands Use the showbf command to see how many nodes are available for use Use the diagnose command to show the partitions Use the check job command to check the status of a particular job 28 28 28 28 29 29 29 29 29 29 29 29 29 29 29 Default File Location u loadl maui bin showg Notes None Copyright 1998 Maui High Performance Computing Center All rights reserved Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved showres showres ARGS lt RESID gt Purpose show detailed reservation information Argument Description g show grep able output with nodename on every line h show usage help n display information regarding all nodes reserved by lt RESID gt 0 display all reservations which overlap lt RESID gt in time r display reservation timeframes in relative time mode S display summary reservation information show verbose output If used with the n flag the command will display all Vy reservations found on nodes contained in lt RESID gt Otherwise it will show long reservation start dates including the reservation year Parameter Description RESID ID of reservation of interest optional Access This command can be run by any Maui administrator or by any valid user if the parameter
210. mmands The log output can also be evaluated to see if any unexpected states were entered Test mode can also be used to locate system problems which need to be corrected Like simulation mode this mode can also be used to safely test drive the scheduler as well as obtain confidence over time of the reliability of the software Once satisfied the scheduling mode can be changed from TEST to NORMAL to begin live scheduling To set up Maui in test mode use the following step gt vi maui cfg change SERVERMODE NORMAL to SERVERMODE TEST gt maui Remember that Maui running in test mode will not interfere with your production scheduler be it Loadleveler PBS or even another version of Maui NOTE If you are running multiple versions of Maui be they in simulation normal or test mode make certain that they each reside in different home directories to prevent conflicts with config and log files statistics checkpointing and lock files Also each instance of Maui should run using a different SERVERPORT parameter to avoid socket conflicts Maui client commands can be pointed to the proper Maui server by using the appropriate command line arguments or by setting the environment variable MAUIHOMEDIR 2 3 1 3 Normal Mode For the adventurous at heart or if you simply have not yet been properly burned by directly installing a large totally new mission critical piece of software or if you are bringing up a new or development system
211. mponent allows a site to prioritize jobs based on political issues such as the relative importance of certain groups or accounts This allows direct political priorities to be applied to jobs The priority calculation for the credential component is Priority CREDWEIGHT USERWEIGHT J gt U SPrIOr ey F GROUPWEIGHT x J gt G gt Priority ACCOUNTWEIGHT J gt A gt Priority QOSWEIGHT aS AP PTO Itty f CLASSWEIGHT J gt C gt Priority All user group account QoS and class weights are specified by setting the PRIORITY attribute of using the respective CFG parameter namely USERCFG GROUPCFG ACCOUNTCFG QOSCFG and CLASSCFG For example to set user and group priorities the following might be used CREDWEIGHT 1 USERWEIGHT 1 GROUPWEIGHT i USERCFG john PRIORITY 2000 USERCFG paul PRIORITY 1000 GROUPCFG staff PRIORITY 10000 Bss or queue priority may also be specified via the resource manager where supported i e PBS queue priorities However if Maui class priority values are also specified the resource manager priority values will be overwritten All priorities may be positive or negative 5 1 2 2 Fairshare FS Component Fairshare components allow a site to favor jobs based on short term historical usage The Fairshare Overview describes the configuration and use of Fairshare in detail After the brief reprieve from complexity found in the QOS factor we come to the Fairshare factor This fact
212. n RESERVATIONCORRUPTION lt MESSAGE gt corruption has been detected lt RESNAME gt lt RESTYPE gt lt NAME gt lt PRESENTTIME gt JA new reservation has MSTARTTIME gt lt ENDTIME gt been created lt NODECOUNT gt lt RESNAME gt lt RESTYPE gt lt PRESENTTIME gt A reservation has lt STARTTIME gt lt ENDTIME gt _ been destroyed lt NODECOUNT gt RESERVATIONCREATED RESERVATIONDESTROYED The interface to the RMFAILURE lt MESSAGE gt resource manager has failed Perhaps the most valuable use of the notify program stems from the fact that additional notifications can be easily inserted into Maui to handle site specific issues To do this locate the proper block routine specify the correct conditional statement and add a call to the routine notify lt MESSAGE gt See Also N A Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 14 5 Issues with Client Commands Client Overview Maui clients are implemented as symbolic links to the executable maui_client When a maui client command is run the client executable determines the name under which it is run and behaves accordingly At the time Maui was configured a home directory was specified The Maui client will attempt to open the config file maui cfg in this home directory on the node where the client command is executed This means that the home directory specified at configure time must be available
213. n time by its specified wall clock limit These fields are empty until an account has completed at least one job Example 2 showstats g Group Statistics Initialized Tue Aug 26 14 32 39 ZSSS Running Completed l GroupName GID Jobs Procs ProcHours Jobs PHReq PHDed FSTgt AvgXF MaxXF AvgQH Effic WCAcc univ 214 16 92 1394 52 229 39 15 18486 45 26 7003 5 41 54 40 00 O47 8 15 5 21 90 70 34 69 daf 204 11 63 859 27 43 7 35 6028 14 76 3448 4 20 45 6 25 0 71 5 40 3 14 98 64 40 83 dnavy 207 6 72 T2812 90 15 38 5974 14 63 3170 7 18 81 6 25 O37 4 88 0 52 82 01 24 14 govt 232 3 24 220 72 TI AS eh6 253 7 6 21 1526 6 9 06 1 53 14 81 0 42 98 73 28 40 asp 227 0 0 0 00 12 2 05 6013 14 72 958 6 569 ZeD 1 78 8 61 5 60 83 64 17 04 derim 229 0 0 0 00 74 12 65 669 1 64 352 5 2 09 0 50 15 93 0 51 96 03 32 60 dchall 274 0 0 0 00 3 0 252 447 1 10 169 2 1 00 25 00 0 52 0 88 2 49 95 82 33 67 nih 239 0 0 0 00 17 2 91 170 0 42 148 1 O68 S S5 0 95 1 83 0 14 97 59 84 31 darmy 205 0 0 0 00 31 5 30 366 0 90 B39 0 32 6 25 0 14 0 59 0 07 81 33 12 73 systems 80 0 0 0 00 6 1 03 67 0 16 22 4 O23 ss4 4 07 8 49 1 23 28 68 37 34 pde 252 0 0 0 00 1 0 17 64 0 16 Sel 0 203 e gt s gt 10 85 10 85 LO LE SOS 9O 7 40 staff I 0 0 0 00 1 0 17 12 0 03 0 2 Oe 00 S 4 0 04 0 04 OL 24 27 1 20 This example shows a statistical listing of all active groups The top line Group Statistics Initialized of the output indicates the beginning o
214. nd record credential based usage statistics Information from these files can be seen via the diagnose f command See Also Simulation Overview SMP Aspects Fairness Policies Prioritization Resource Allocation Policies Shared vs Dedicated SMP nodes are often used to run jobs which do not use all available resources on that node How Maui handles these unused resources is controlled by the parameter NODEACCESSPOLICY If this parameter is set to SHARED Maui will allow tasks of other jobs to use the resources If this parameter is set to DEDICATED Maui will mark these resources unavailable for use by other jobs Reservations Diagnosing System Behavior Problems Maui provides a number of commands for diagnosing system behavior Scheduling in a complicated task and oftentimes a scheduler will behave exactly as you said which may not be exactly what you want Diagnosing thus includes both looking for system failures as well as determining current functioning system behavior Quite often problems may be corrected through configuration changes which more accurately reflect a site s desires When diagnosing system problems the diagnose command may become your best friend This command provides detailed information about scheduler state and also performs a large number of internal sanity checks presenting problems it finds as warning messages Currently the diagnose command provide in depth analysis of the following objects and
215. nding and administrative reservations Maui can also create priority reservations These reservations are used to allow the benefits of out of order execution such as is available with backfill without the side effect of job starvation Starvation can occur in any system where the potential exists for a job to be overlooked by the scheduler for an indefinate period In the case of backfill small jobs may continue to be run on available resources as they become available while a large job sits in the queue never able to find enough nodes available simultaneously to run on To avoid such situations priority reservations are created for high priority jobs which cannot run immediately When making these reservations the scheduler determines the earliest time the job could start and then reserves these resources for use by this job at that future time By default only the highest priority job will receive a priority reservation However this behavior is configurable via the RESERVATIONDEPTH policy Maui s default behavior of only reserving the highest priority job allows backfill to be used in a form known as liberal backfill This liberal backfill tends to maximize system utilization and minimize overall average job turnaround time However it does lead to the potential of some lower priority jobs being indirectly delayed and may lead to greater variance in job turnaround time The RESERVATIONDEPTH parameter can be set to a very large value essentially
216. ne processor 512MB or memory and 2 GB of local disk A key aspect of a task is that the resources associated with the task must be allocated as an atomic unit without spanning node boundaries A task requesting 2 processors cannot be satisfied by allocating 2 uniprocessor nodes nor can a task requesting 1 processor and 1 GB of memory be satisfied by allocating 1 processor on one node and memory on another In Maui when jobs or reservations request resources they do so in terms of tasks typically using a task count and a task definition By default a task maps directly to a single processor within a job and maps to a full node within reservations In all cases this default definition can be overridden by specifying a new task definition Within both jobs and reservations depending on task definition it is possible to have multiple tasks from the same job mapped to the same node For example a job requesting 4 tasks using the default task definition of 1 processor per task can be satisfied by two dual processor nodes 3 2 1 7 PE The concept of the processor equivalent or PE arose out of the need to translate multi resource consumption requests into a scalar value It is not an elementary resource but rather a derived resource metric It is a measure of the actual impact of a set of requested resources by a job on the total resources available system wide It is calculated as PE MAX ProcsRequestedByJob TotalConfiguredProcs
217. nforced as a series of weekly reservations which only cover the specified timeframe The SRFEATURES parameter indicates that each of these nodes must have the node feature Largememory configured As described above SRMAXTIME indicates that jobs using this reservation can only use it for one hour What does this mean It means the job and the reservation can only overlap for one hour Clearly jobs requiring an hour or less of wallclock time meet this constraint However so does a four hour job that starts on Monday at 5 00 AM or a 12 hour job which starts on Friday at 4 00 PM Also note the SRTIMELOGIC setting It is set to AND This means that jobs must not only meet the SRMAXTIME access constraint but must also meet one or more of the other access constraints In this example the job can use this reservation if it can utilize the access specified via SRQOSLIST or SRACCOUNTLIST i e it is assigned a QOS of high low or special or the submitter of the job has an account which satisfies the projectX and projectyY criteria More on this below NOTE See the QOS Overview for more info about QOS configuration and usage 7 1 5 2 4 Affinity One aspect of reservations that has not yet been discussed is something called reservation affinity By default jobs gravitate towards reservations in a behavior known as positive affinity This allows jobs to run on the most constrained resources leaving other unreserved resources free for use by o
218. ng Maui to act as a simple FIFO Once the summed component weight is determined this value is then bounded resulting in a priority ranging between 0 and MAX_PRIO_VAL which is currently defined as 1000000000 one billion In no case will a job obtain a priority in excess of MAX_PRIO_VAL through its priority subcomponent values Using the setspri command site admins may adjust the base calculated job priority by either assigning a relative priority adjust or an absolute system priority A relative priority adjustment will cause the base priority to be increased or decreased by a specified value Setting an absolute system priority SPRIO will cause the job to receive a priority equal to MAX_PRIO_VAL SPRIO and thus guaranteed to be of higher value than any naturally occurring job priority Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 5 1 2 Job Priority Factors Maui allows jobs to be prioritized based on a range of job related factors These factors are broken down into a two level hierarchy of priority factors and subfactors each of which can be independently assigned a weight This approach provides the administrator with detailed yet straightforward control of the job selection process The table below highlights the components and subcomponents which make up the total job priority Component SubComponent Metric Bees denad user user specific priority See USERCFG group specific priority S
219. ng and using these reservations 7 1 1 6 Reservation Behavior As mentioned above a given reservation may have one or more access criteria A job can utilize the reserved resources if it meets at least one of these access criteria It is possible to stack multiple reservations on the same node In such a situation a job can only utilize the given node if it meets at least access criteria of each active reservation on the node 7 1 1 7 Other Reservation Attributes Charge Account Allows a reservation to charge for resources which are dedicated to the reservation but not used by any job See also N A Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 7 1 2 Administrative Reservations Administrative reservations behave much like standing reservations but are generally created to address non periodic one time issues All admin reservations are created using the setres command and are persistent until they expire or are removed using the releaseres command See also Reservation Overview Backfill Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 7 1 3 Standing Reservations Standing reservations build upon the capabilities of advance reservations to enable a site to enforce advanced usage policies in an efficient manner Standing reservations provide a superset of the capabilities typically found in a batch queuing system s class or queue a
220. nitor and control the job Example 2 PBS command file PBS W X NODESET ONEOF NETWORK DMEM 64 Job will have resources allocated subject to network based nodeset constraints Further each task will dedicate 64 MB of memory Example 3 qsub l nodes 4 walltime 1 00 00 W x FLAGS ADVRES john 1 Job will be forced to run within the john 1 reservation See Also Resource Manager Overview Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved L__ 13 4 Adding New Resource Manager Interfaces Maui currently interfaces with about 6 different resource manager systems Some of these interact through a resource manager specific interface ie OpenPBS PBSProc Loadleveler while others interact through a simple text based interfaces known as Wiki see the Wiki Overview For most resource managers either route is possible depending on where it is easiest to focus development effort Use of Wiki generally requires modifications on the resource manager side while creation of a new resource manager specific Maui interface would require more changes to Maui mods If a scheduling API already exists within the resource manager creation a a resource manager specific Maui interface is often selected Regardless of the interface approach selected adding support for a new resource manager is typically a straight forward process for about 95 of all supported features The final 5 of features usually re
221. nts of their subsequent jobs may be very different from their initial run Users often do not update their batch command files even though these constraints may be unnecessarily limiting the resources available to their jobs for two reasons 1 users do not know how much their performance will improve if better information were provided 2 users do not no exactly what resources their jobs are utilizing and are afraid to lower their job s resource requirements since doing so might cause their job to fail To help with determining accurate per job resource utilization information Maui provides the FEEDBACKPROGRAM facility This tool allows sites to send detailed resource utilization information back to users via email to store it in a centralized database for report preparation or use it in other ways to help users refine their batch jobs Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 15 2 User Level Statistics Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 15 3 Enhancing Wallclock Limit Estimates Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 15 4 Providing Resource Availability Information Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 15 5 Job Start Time Estimates Under Construction
222. ob Attdhut s A queue may be associated with a default job duration default size or default resource requirements Hos Condes R clase may constrain job execution to a particular set A queue may constrain the attributes of jobs which may submitted including setting limits such as max wallclock time minimum number of processors etc ee ace Tat A queue may constrain who may submit jobs into it based on user lists group lists etc Special Access A queue may associate special privileges with jobs including adjusted job priority As stated previously most resource managers allow full class configuration within the resource manager Where additional class configuration is required the CLASSCKFG parameter may be used Job Constraints Maui tracks class usage as a consumable resource allowing sites to limit the number of jobs using a particular class This is done by monitoring class initiators which may be considered to be a ticket to run in a particular class Any compute node may simultaneously support serveral types of classes and any number of initiators of each type By default nodes will have a one to one mapping between class initiators and configured processors For every job task run on the node one class initiator of the appropriate type is consumed For example a 3 processor job submitted to the class batch will consume three batch class initiators on the nodes where it is run Using queues as consumable resources allows
223. ob or per QOS basis to constrain which resources a job may have access to See the QOS Overview for more information By default QOS s and jobs allow global partition access If no partition is specified Maui creates a single partition named DEFAULT into which all resources are placed In addition to the DEFAULT partition a pseudo partition named ALL is created which contains the aggregate resources of all partitions NOTE While DEFAULT is a real partition containing all resources not explicitly assigned to another partition the ALL partition is only a convenience construct and is not a real partition thus it cannot be requested by jobs or included in configuration ACL s 7 2 1 Defining Partitions 7 2 2 Managing Partition Access 7 2 3 Requesting Partitions 7 2 4 Miscellaneous Partition Issues 7 2 1 Defining Partitions Node to partition mappings are established using the NODECFG parameter in Maui 3 0 7 and higher as shown in the example below NODECFG node001 PARTITION ast ronomy NODECFG node002 PARTITION ast ronomy NODECFG node049 PARTITION math In earlier versions of Maui node to partition mappings were handled in the machine config file machine cfg using the PARTITION keyword as in the example below node001 PARTITION astronomy node002 PARTITION astronomy node049 PARTITION math However if using partitions it is HIGHLY recommended that Maui 3 0 7 or higher be used 7 2 2 Managing Partition
224. obs multiple steps occur 3 3 1 5 Backfill Jobs 3 3 1 6 Update Statistics 3 3 1 7 Handle User Requests User requests include any call requesting state information configuration changes or job or resource manipulation commands These requests may come in the form of user client calls peer daemon calls or process signals 3 3 1 8 Perform Next Scheduling Cycle Maui operates on a polling event driven basis When all scheduling activities are complete Maui will process user requests until a new resource manager event is received or an internal event is generated Resource manager events include activities such as a new job submission or completion of an active job addition of new node resources or changes in resource manager policies Internal events include admin schedule requests reservation activation deactivation or the expiration of the RMPOLLINTERVAL timer 3 3 2 Detailed Job Flow 3 3 2 1 Determine Basic Job Feasibility The first step in scheduling is determining which jobs are feasible This step eliminates jobs which have job holds in place invalid job states 1 e Completed Not Queued Defered etc or unsatisfied preconditions Preconditions may include stage in files or completion of preliminary job steps 3 3 2 2 Prioritize Jobs With a list of feasible jobs created the next step involves determining the relative priority of all jobs within that list A priority for each job is calculated based on job attr
225. ok again at the system statistics gt showstats Note that a few more fields are filled in now that some jobs have completed providing information on which to generate statistics Advance the scheduler 2 more steps gt schedctl S 2I The 2I argument indicates that the scheduler should advance 2 steps and that it should Ignore user input until it gets there This prevents the possibility of obtaining showq output from iteration 5 rather than iteration 6 gt showg r It looks like the 5 processor job completed as expected while another 20 processor job completed early The scheduler was able to start another 20 processor job and five serial jobs to again utilize all idle resources Don t worry this is not a stacked trace designed to make the Maui scheduler appear omniscient We have just gotten lucky so far and have the advantage of a deep default queue of idle jobs Things will get worse Let s look at the idle workload more closely gt showg i This output is listed in priority order We can see that we have a lot of jobs from a small group of users many larger jobs and a few remaining easily backfillable jobs let s step a ways through time To speed up the simulation let s decrease the default LOGLEVEL to avoid unnecessary logging gt changeparam LOGLEVEL 0 changeparam can be used to immediately change the value of any parameter The change is only made
226. on all hosts where the maui client commands will be executed This also means that a maui cfg file must be available in this directory When the clients open this file they will try to load the MAUISERVER and MAUIPORT parameters to determine how to contact the Maui server NOTE The home directory value specified at configure time can be overridden by creating an etc maui cfg file or by setting the MAUIHOMEDIR environment variable Once the client has determined where the Maui server is located it creates a message adds an encrypted checksum and sends the message to the server Note that the Maui client and Maui server must use the same secret checksum seed for this to work When the Maui server receives the client request and verifies the checksum it processes the command and returns a reply Diagnosing Client Problems The easiest way to determine where client failures are occuring is to utilize built in maui logging On the client side use the L flag For example gt showq L9 NOTE Maui 3 0 6 and earlier specified the desired client side logging level using the D flag 1 e showq D 9 This will dump out a plethora of information about loading the configfile connecting to the maui server sending a request and receiving a response Wading through this information almost always will reveal the source of the problem If it does not the next step is to look at the maui server side logs The easiest way to do this is to u
227. onditional breakpoints which make debugging much easier For debuggers which do not have such capabilities the TRAP parameters are of value allowing breakpoints to be set which only trigger when specific routines are processing particular nodes jobs or reservations See the TRAPNODE TRAPJOB TRAPRES and TRAPFUNCTION parameters for more information Controlling behavior after a crash Setting CRASHMODE See also Troubleshooting Individual Jobs Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 9 3 Profiling Current and Historical Usage Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 9 4 Testing New Versions and Configurations Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 9 5 Answering What If Questions with the Simulator Under Construction see 16 0 Simulations Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 10 0 Managing Shared Resources SMP Issues and Policies 10 1 Consumable Resource Handling 10 2 Load Balancing Features Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 10 1 Consumable Resource Handling Maui is designed to inherently handle consumable resources Nodes possess resources and workload jobs consume resources Maui tracks
228. onfiguration would be a good initial stab maui cfg reserve 16 processors during primetime for jobs requiring less than 2 hours to complete SRNAME 0 fast SRTASKCOUNT 0 16 SRDAYS 0 MON TUE WED THU FRI SRSTARTTIME 0 8 00 00 SRENDTIME 0 17 00 00 SRMAXTIME 0 2 00 00 prioritize jobs for Fairshare XFactor and Resources RESOURCEWEIGHT 20 XFACTORWEIGHT 100 FAIRSHAREWEIGHT 100 disable SMP node sharing NODEACCESSPOLICY DEDICATED Group Meterology FSTARGET 45 Group Statistics FSTARGET 35 Monitoring The command diagnose f will allow you to monitor the effectiveness of the fairshare component of your job prioritization Adjusting the Fairshare priority factor up or down will make fairshare more less effective Note that a tradeoff must occur between fairshare and other goals managed via job prioritization diagnose p will help you analyze the priority distributions of the currently idle jobs The showgrid AVGXFACTOR command will provide a good indication of average job turnaround while the profiler command will give an excellent analysis of longer term historical performance statistics Conclusions Any priority configuration will need to be tuned over time because the effect of priority weights is highly dependent upon the site specific workload Additionally the priority weights themselves are part of a feedback loop which adjust the site workload However most sites quickly stabilize an
229. ons do have value it is important to note that within Maui the standing reservation facility provides significantly improved flexibility and should be used in the vast majority of cases where partitions are required under other resource management systems Standing reservations provide time flexibility improved access control features and more extended resource specification options Also another Maui facility called Node sets allows intelligent aggregation of resources to improve per job node allocation decisions In cases where system partitioning is considered for such reasons node sets may be able to provide a better solution Still one key advantage of partitions over standing reservations and node sets is the ability to specify partition specific policies limits priorities and scheduling algorithms although this feature is rarely required An example of this need may be a cluster consisting of 48 nodes owned by the Astronomy Department and 16 nodes owned by the Mathematics Department Each department may be willing to allow sharing of resources but wants to specify how their partition will be used As mentioned earlier many of Maui s scheduling policies may be specified on a per partition basis allowing each department to control the scheduling goals within their partition The partition associated with each node must be specified as indicated in the Node Location section With this done partition access lists may be specified on a per j
230. optimistic scheduling and development job support Example QOSCFG high FLAGS PREEMPTOR QOSCFG med OQOSCFG low FLAGS PREEMPTEER See Also N A Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 9 0 Evaluating System Performance Statistics Profiling Testing and Simulation 9 1 Maui Performance Evaluation Overview 9 2 Job and System Statistics 9 3 Profiling Current and Historical Usage 9 4 Testing New Versions and Configurations 9 5 Answering What If Questions with the Simulator Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ CUELA 9 1 Maui Performance Evaluation Overview Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 9 2 Job and System Statistics Maui maintains a large number of statistics and provides several commands to allow easy access to and helpful consolidation of this information These statistics are of three primary types mi 9 2 2 Profiling Historical Usage 9 2 3 FairShare Usage Statistics 9 2 1 Real Time Statistics Mavi provides real time statistical information about how the machine is running from a scheduling point of view The showstats commands is actually a suite of commands providing detailed information on an overall scheduling basis as well as a per user group account and node basis This command gets its information from
231. or is used to adjust a job s priority based on the historical percentage system utilization of the jobs user group account or QOS This allows you to steer the workload toward a particular usage mix across user group account and QOS dimensions The fairshare priority factor calculation is Priority FSWEIGHT MIN FSCAP FSUSERWEIGHT DeltaUserFSUsage FSGROUPWEIGHT DeltaGroupFSUsage FSACCOUNTWEIGHT DeltaAccountFSUsage FSQOSWEIGHT DeltaQOSFSUsage FSCLASSWEIGHT DeltaClassFSUsage All WEIGHT parameters above are specified on a per partition basis in the maui cfg file The Delta Usage components represents the difference in actual fairshare usage from a fairshare usage target Actual fairshare usage is determined based on historical usage over the timeframe specified in the fairshare configuration The target usage can be either a target floor or ceiling value as specified in the fairshare config file The fairshare documentation covers this in detail but an example should help obfuscate things completely Consider the following information associated with calculating the fairshare factor for job X Job X User A Group B Account C QOS D Class E User A Fairshare Target 50 0 Current Fairshare Usage 45 0 Group B Fairshare Target NONE Current Fairshare Usage 65 0 Account C Fairshare Target 25 0 Current Fairshare Usage 35 0 QOS 3 Fairshare Target 10 0 Current Fairshare Usage 25 0 Class
232. ority Management ad 11 3 Suspend Resume Handlin A 11 4 Checkpoint Restart Facilities Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved L 11 1 Job Holds Holds and Deferred Jobs A job hold is a mechanism by which a job is placed in a state where it is not eligible to be run Maui supports job holds applied by users admins and even resource managers These holds can be seen in the output of the showq and checkjob commands A job witha hold placed on it cannot be run until the hold is removed If a hold is placed on a job via the resource manager this hold must be released by the resource manager provided command i e Ilhold for Loadleveler or qhold for PBS Maui supports two other types of holds The first is a temporary hold known as a defer A job is deferred if the scheduler determines that it cannot run This can be because it asks for resources which do not currently exist does not have allocations to run is rejected by the resource manager repeatedly fails after start up etc Each time a job gets deferred it will stay that way unable to run for a period of time specified by the DEFERTIME parameter If a job appears with a state of deferred it indicates one of the previously mentioned failures has occurred Details regarding the failure are available by issuing the checkjob lt JOBID gt command Once the time specified by DEFERTIME has elapsed the job is automatically released and
233. orms while consumed tracking is used in shared SMP environments Using the selected fairshare usage metric Maui continues to update the current fairshare window until it reaches a fairshare window boundary at which point it rolls the fairshare window and begins updating the new window The information for each window is stored in its own file located in the Maui statistics directory Each file is named FS lt EPOCHTIME gt where lt EPOCHTIMES is the time the new fairshare window became active Each window contains utilization information for each entity as well as for total usage A sample fairshare data file is shown below Fairshare Data File Duration 172800 Seconds Fri Aug 18 18 00 00 Starting User USERA 150000 000 User USERB 150000 000 User USERC 200000 000 User USERD 100000 000 Group GROUPA 350000 000 Group GROUPB 250000 000 Account ACCTA 300000 000 Account ACCTB 200000 000 Account ACCTC 100000 000 QOS 0 50000 000 QOS i 450000 000 QOS 2 100000 000 TOTAL 600000 00 Note that the total processor hours consumed in this time interval is 600 000 processor seconds Since every job in this example scenario had a user group account and QOS assigned to it the sum of the usage of all members of each category should equal the total usage value i e USERA USERB USERD GROUPA GROUPB ACCTA ACCTC QOSO QOS2 TOTAL When Maui needs to determine current fairshare usage for a particular entity i
234. ounting statistics for the system Historical statistics cover the timeframe from the most recent execution of the resetstats command Example 1 o showstats a Account Statistics Initialized Tue Aug 26 14 32 39 Running Completed Account Jobs Procs ProcHours Jobs PHReq PHDed FSTgt AvgXF MaxXF AvgQH Effic WCAcc 137651 16 92 1394 52 229 39 15 18486 45 26 7003 5 41 54 40 00 0 77 8 15 5 21 90 70 34 69 462212 11 63 855 27 43 7 35 6028 14 76 3448 4 20 45 6 25 0 71 5 40 3 14 98 64 40 83 462213 6 72 728 12 90 15 38 5974 14 63 3170 7 18 81 6 25 0 37 4 88 0 52 82 01 24 14 005810 3 24 220 72 77 13 16 2537 6 21 1526 6 9 06 1 53 14 81 0 42 98 73 28 40 175436 0 0 0 00 12 2 05 6013 14 72 958 6 5 69 2 50 1 78 8 61 5 60 83 64 17 04 000102 0 0 0 00 1 0 17 64 0 16 Bec 0 03 10 85 10 85 10 77 27 90 7 40 000023 0 0 0 00 1 0 17 12 0 03 0 2 0 00 0 04 0 04 019 21 21 1 20 This example shows a statistical listing of all active accounts The top line Account Statistics Initialized of the output indicates the beginning of the timeframe covered by the displayed statistics The statistical output is divided into two categories Running and Completed Running statistics include information about jobs that are currently running Completed statistics are compiled using historical information from both running and completed jobs The fields are as follows Account Jobs Procs ProcHours Jobs PHReq PH
235. ources are specified on a per task basis and currently include processors local disk real memory and swap The access control list supported for standing reservations includes users groups accounts job classes and QOS levels Standing reservations can be configured to be permanent or periodic on a daily or weekly basis and can accept a daily or weekly start and end time Regardless of whether a standing reservation recurs on a daily or weekly basis standing reservations are enforced using a series of reservations extending a number of periods into the future as controlled by the SRDEPTH parameter For example the following configuration will create a standing reservation for 6 processors and 3 GB of memory for use by the interactive class during business hours SRNAME 0 interactive SRTASKCOUNT 0 6 SRRESOURCES 0 PROCS 1 MEM 512 SRPERIOD 0 DAY SRDAYS 0 MON TUE WED THU FRI SRSTARTTIME 0 9 00 00 SRENDTIME 0 17 00 00 SRCLASSLIST 0 interactive In Maui 3 2 0 or later this could be accomplished using the SRCEG parameter as in the example below SRCFG interactive STARTTIME 9 00 00 ENDTIME 17 00 00 SRCFG interactive PERIOD DAY DAYS MON TUE WED THU FRI SRCFG interactive TASKCOUNT 6 RESOURCES PROCS 1 MEM 512 SRCFG interactive CLASSLIST interactive Let s examine the new parameters one at a time SRNAME simply gives the standing reservation a name for reference by Maui commands It is not required but mak
236. ow note that job X which meets access criteria for both reservation A and B allocates a portion of its resources from each reservation and the remainder from resources outside of both reservations Although by default reservations make resources available to jobs which meet particular criteria Maui can be configured to constrain jobs to only run within accessible reservations This can be requested by the user on a job by job basis using a resource manager extension flag or can be enabled administratively via a QoS flag For example assume two reservations were created as shown below gt setres g staff d 8 00 00 node 1 4 reservation staff 1 created on 4 nodes gt setres u john tasks reservation john 1 created on two nodes If the user john who happened to also be a member of the group staff wanted to force his job to run within a particular reservation he could do so using the FLAGS resource manager extension Specifically in the case of a PBS job the following submission would force the job to run within the staff 1 reservation gt qsub 1 nodes 1 walltime 1 00 00 W X FLAGS ADVRES staff 1 testjob cmd Note that for this to work PBS will need to have resource manager extensions enabled as described in the PBS Resource Manager Extension Overview If the user simply wants the job to run on reserved resources but does not care which he could submit the job with gt qsub 1 nodes 1 walltime 1 00 00 W
237. pecifies how Maui should LASTAVAILABLE MINRESOURCE allocate available resources to NODEALLOCATIONPOLICY MINRESOURCE NODEALLOCATIONPOLICY CPULOAD MACHINEPRIO LOCAL LASTAVAILABLE jobs See the Node Allocation Maui will apply the node allocation policy CONTIGUOUS MAXBALANCE or section of the Admin manual for MINRESOURCE to all jobs by default FASTEST more information list of space delimited lt ATTR gt lt VALUE gt specifies node specific attributes pairs where lt ATTR gt is one of the for the node indicated in the NODECFG nodeA MAXJOB 2 SPEED 1 2 following array field See the Node NODECFG X ACCESS MAXJOB NONE Configuration Overview for Maui will only only two simultaneous jobs to run on MAXJOBPERUSER MAXLOAD more information NOTE this node nodeA and will assign a relative machine speed FRAME SLOT SPEED PROCSPEED parameter is enabled in Maui of 1 2 to this node PARTITION NODETYPE FEATURES 3 0 7 and higher length of time Maui will assume down drained offline or NODEDOWNSTATEDELAYTIME 0 30 00 corrupt nodes will remain Maui will assume down drained and corrupt nodes unavailable for scheduling ifa are not available for scheduling for at least 30 minutes NODEDOWNSTATEDELAYTIME DD HH MM SS 0 00 00 system reservation is not from the current time Thus these nodes will never be explicitly created for the node allocated to starting jobs Also these nodes will only NOTE This parameter is be available for rese
238. per user group account and QOS fairshare configuration fs cfg by default In earlier versions of Maui Maui 3 0 6 and earlier fairshare configuration information was specified via the fs cfg file In Maui 3 0 7 and higher although use of the fs cfg file is still supported it is recommended that the CFG suite of parameters ACCOUNTCEG CLASSCFG GROUPCFG QOSCKEG and USERCFG be used Both approaches allow specification of per user group account and QOS fairshare in terms of target limits and target types As Maui runs it records how available resources are being utilized Each iteration RMPOLLINTERVAL seconds it updates fairshare resource utilization statistics Currently resource utilization is measured in accordance with the FSPOLICY parameter allowing various aspects of resource consumption information to be tracked This parameter allows selection of both the types of resources to be tracked and the method of tracking It provides the option of tracking usage by dedicated or consumed resources where dedicated usage tracks what the scheduler assigns to the job while consumed usage tracks what the job actually uses An example may clarify this Assume a 4 processor job is running a parallel bin sleep for 15 minutes It will have a dedicated fairshare usage of 1 proc hour but a consumed fairshare usage of essentially nothing since it did not consume anything Most often dedicated fairshare usage is used on dedicated resource platf
239. ple the XFactors of two jobs were averaged to obtain an average XFactor of 1 24 for jobs requiring over 2 hours 8 minutes but not more than 4 hours 16 minutes and between 5 and 8 nodes Totals along the bottom provide overall XFactor averages weighted by job node and node seconds Related Commands None Default File Location u loadl maui bin showgrid Notes None Copyright 1998 Maui High Performance Computing Center All rights reserved Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ showq showg i r p PARTITION h Purpose Shows information about running idle and non queued jobs Permissions This command can be run by any user However the i and r flags can only be used by Maui Scheduler Administrators Parameters PARTITION Partition number that you wish to inspect Flags h Help for this command 1 Used by Maui Scheduler Administrators to display idle jobs only p Inspect partition specified with PARTITION parameter r Used by Maui Scheduler Administrators to display running jobs only Description Since LoadLeveler is not actually scheduling jobs the job ordering it displays is no longer valid The showq command displays the actual job ordering under the Maui Scheduler When used without flags this command displays all jobs in active idle and non queued states Example 1 o showq ACTIVE JOBS
240. pyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 3 3 Scheduling Iterations and Job Flow 331 Scheduling Iterations 3 3 1 1 Update State Information 3 3 1 2 Refresh Reservations 3 3 1 3 Schedule Reserved Jobs 3 3 1 4 Schedule Priority Jobs 3 3 1 5 Backfill Jobs 3 3 1 6 Update Statistics 3 3 1 7 Handle User Requests oe ae a ee ee ee ee 3 3 1 8 Perform Next Scheduling Cycle 3 3 2 Detailed Job Flow L 3 3 2 1 Determine Basic Job Feasibility 3 3 2 2 Prioritize Jobs 3 3 2 3 Enforce Configured Throttling Policies 3 3 2 4 Determine Resource Availability 3 3 2 5 Allocate Resources to Job 3 3 2 6 Distribute Jobs Tasks Across Allocated Resources oe a a ee 3 3 2 7 Launch Job 3 3 1 Scheduling Iterations In any given scheduling iteration many activities take place These are broken into the following major categories Update State Information Refresh Reservations Schedule Reserved Jobs Schedule Priority Jobs Backfill Jobs Update Statistics Handle User Requests 3 3 1 1 Update State Information Each iteration the scheduler contacts the resource manager s and requests up to date information on compute resources workload and policy configuration On most systems these calls are to a centralized resource manager daemon which possesses all information 3 3 1 2 Refresh Reservations 3 3 1 3 Schedule Reserved Jobs 3 3 1 4 Schedule Priority Jobs In scheduling j
241. quires a bit more effort as each resource manager has a number of distinctive and unique concepts which must be addressed 13 4 1 Resource Manager Specific Interfaces 13 4 2 Wiki Interface 13 4 1 Resource Manager Specific Interfaces If the resource manger specific interface is desired then typically a scheduling API library header file combo is required i e for PBS libpbs a pbs_ifl h etc This resource manager provided API provides calls which can be linked into Maui to obtain the raw resource manager data including both jobs and compute nodes Additionally this API should provide policy information about the resource manager configuration if it is desired that such policies be specified via the resource manager rather than the scheduler and that Maui know of and respect these policies The new lt X gt Interface c module would be responsible for loading information from the resource manager translating this information and then populating the appropriate Maui data structures The existing LLInterface c PBSInterface c and WikiInterface c modules provide templates indicating how to do this The first step in this process is defining the new resource manager type This is accomplished by modifying maui_struct h and maui_global h header files to define the new RMTYPE parameter value With this defined the RMInterface c module must be modified to call the appropriate resource manager specific calls which will eventually be creat
242. r GREEDY backfill algorithms one of the following FIRSTFIT specifies what backfill algorithm BACKFILLPOLICY BESTFIT GREEDY or NONE FIRSTFIT will be used BACKFILLPOLICY BESTFIT one of DEBITALLWC specifies how Maui should BANKCHARGEPOLICY DEBITALLWC DEBITSUCCESSFULWC charge a completed job agamst Maui will charge an account for the resources BANKCHARGEPOLICY DEBITSUCCESSFULCPU or DEBITSUCCESSFULWC jan allocation manager See the dedicated to a job regardless of how well the job uses DEBITSUCCESSFULPE Allocation Manager OvervieW these resources and regardless of whether or not the for details job completes successfully specifies whether or not Maui BANKDEFERJOBONFAILURE ON or OFF OFF should defer jobs if the BANKDEFERJOBONFAILURE ON allocation bank is failing BANKFALLBACKACCOUNT lt STRING gt NONE pecount to use if eee BANKFALLBACKACCOUNT bottomfeeder account is out of allocations BANKPORT lt INTEGER gt 40560 co to use to contact allocation laa NKpoRT 40555 manager bank name of host on which BANKSERVER lt STRING gt NONE allocation manager bank service JBANKSERVER zephyrl resides number of seconds Maui will BANKTIMEOUT lt INTEGER gt 9 wait before timing out on a bank BANKTIMEOUT 00 00 30 connection BANKTYPE one of QBANK RESD or FILE NONE oe type of allocation aNKTYPE QBANK specifies the weight to be BYPASSWEIGHT lt INTEGER gt 0 applied to a job s backfill bypass vpagcumIGHT 5000 count when determining a job s prio
243. r all resource usage based on utilized CPU time cancel jobs which exceed specified resource limits notify users of job cancellation due to resource utilization limit violations The following Maui config file should work maui cfg allow jobs to share node NODEACCESSPOLICY SHARED track background load NODELOADPOLICY ADJUSTPROCS NODEUNTRACKEDLOADFACTOR he 2 favor short jobs disfavor large jobs QUEUET IMEWEIGHT 0 RESOURCEWEIGHT 10 PROCWEIGHT 128 MEMWEIGHT I XFACTOR 1000 disable priority reservations for the default QOS QOSFLAGS 0 NORESERVATION debit by CPU BANKTYPE QBANK BANKSERVER developl BANKPORT 2334 BANKCHARGEMODE DEBITSUCCESSFULLCPU kill resource hogs RESOURCEUTILIZATIONPOLICY ALWAYS RESOURCEUTILIZATIONACTION CANCEL notify user of job events NOTIFYSCRIPT tools notify pl Monitoring The most difficult aspects of this environment are properly reserving space for the untracked background load Since this load is outside the viewing control of the scheduler resource manager there are no constraints on what it can do It could instant grow and overwhelm the machine or just as easily disappear The parameter NODEUNTRACKEDLOADFACTOR provides slack for this background load to grow and shrink However since there is now control over the load the effectiveness of this parameter will depend on the statistical behavior of this load The greater the value the more slack provided
244. r can change over time It is possible that one job can be at the top of the priority queue for a time and then get bypassed by another job submitted later The parameter RESERVATIONPOLICY allows a site to determine what how existing reservations should be handled when new reservations are made The value HIGHEST will cause that all jobs which have ever received a priority reservation will maintain that reservation until they run even if other jobs later bypass them in priority value The value CURRENTHIGHEST will cause that only the current top lt RESERVATIONDEPTH gt priority jobs will receive reservations If a job had a reservation but has been bypassed in priority by another job so that it no longer qualifies as being amongst the top lt RESERVATIONDEPTHS jobs it will lose its reservation Finally the value NEVER indicates that no priority reservations will be made QOS based reservation depths can be enabled via the RESERVATIONOQOSLIST parameter This parameter allows varying reservation depths to be associated with different sets of job QoS s For example the following configuration will create two reservation depth groupings RESERVATIONDEPTH 0 8 RESERVATIONQOSLIST 0 highprio interactive debug RESERVATIONDEPTH 1 2 RESERVATIONQOSLIST 1 batch This example will cause that the top 8 jobs belonging to the aggregate group of highprio interactive and debug QoS jobs will receive priority reservations Additionally the top 2 batch
245. r of MAX_MATTR MAX_ATTR in Maui 3 0 7 and earlier mourli ias A MAX MCLASS msched common h 16 64 node features Maui can track total number of distinct job classes queues available maximum total number of idle active jobs Maui can see and process MAX MJOB maui h maximum number of compute nodes Maui can see and process MAX MNODE MAX_NODE in Maui 3 0 6 and earlier maximum number of partitions supported total number of distinct QOS objects available to jobs total number of distinct reservations allowed per node MAX MPARTITION MAX_PARTITION in Maui maui h 4 3 0 and earlier MAX MQOS maui h 128 MAX MRES DEPTH maui h 5 e maui h msched common h 1560 10000 Maui currently possesses hooks to allow sites to create local algorithms for handling site specific needs in several areas The contrib directory contains a number of sample local algorithms for various purposes The Local c module incorporates the algorithm of interest into the main code The following scheduling areas are currently handled via the Local c hooks Local Job Attributes Local Node Allocation Policies Local Job Priorities Local Fairness Policies total number of distinct standing reservations available 16 128 5 MAX_SRESERVATION total number of tasks allowed per job MAX MTASK MAX_TASK in Maui 3 2 0 and earlier Overview of
246. r second usage of each credential is updated each scheduling iteration decreasing as job s approach their completion time MAXPS 720000 Limits the total number of dedicated processor equivalents which can be MAXPE 128 allocated by active jobs at any given time of processor MAXPE equivalents Limits the number of outstanding seconds a credential may have associated with active jobs It behaves identically to the MAXPS limit job duration above only lacking LDDD HH MM SS the processor MAXWC MAXWC 72 00 00 outstanding second usage of each credential is also updated each scheduling iteration limits the total number of compute nodes which canbe MAXNODE 64 in use by active jobs at any given time Limits the total amount of dedicated memory in MB which can be allocated by a credential s active jobs at any given time MAXNODE of nodes MAXMEM total memory in MB MAXMEM 2048 The example below demonstrates a simple limit specification USERCFG DEFAULT MAXJOB 4 USERCFG john MAXJOB 8 This example will allow user john to run up to 8 jobs while all other users may only run up to 4 Simultaneous limits of different types may be applied per credential and multiple types of credential may have limits specified The next example demonstrates this mixing of limits and is a bit more complicated USERCFG steve MAXJOB 2 MAXNODE 30
247. racy varies from site to site but the site average is rarely better than 40 Because the quality of the walltime estimate provided by the user is so low job reservations for high priority jobs are often later than they need to be So is backfill worth it The short answer is absolutely The longer answer is Qe Di Si OdU et l y Although there do exist some minor drawbacks with backfill its net performance impact on a site s workload is very positive Its like the phrase a rising tide lifts a ships Although a few of the highest priority jobs may get minorly and temporarily delayed they probably got to their position as highest priority as soon as they did because jobs in front of them got to run earlier due to backfill Studies have shown that only a very small fraction of jobs are truly delayed and when they are it is only by a fraction of their total queue time At the same time many jobs are started significantly earlier than would have occurred without backfill Regarding the other problems described don t vorry ve have vays of handling dem 8 2 2 Backfill Algorithm The algorithm behind Maui backfill scheduling is mostly straightforward although there are a number of issues and parameters of which you should be aware First of all Maui makes two backfill scheduling passes For each pass Maui selects a list of jobs which are eligible for backfill On the first pass only those jobs which meet the constraints of the soft fai
248. rchitecture For example queues can be used to allow only particular types of jobs access to certain compute resources Also some batch systems allow these queues to configured so that they only allow this access during certain times of the day or week Standing reservations allow these same capabilities but with greater flexibility and efficiency than is typically found in a normal queue management system Standing Reservations provide a mechanism by which a site can dedicate a particular block of resources for a special use on a regular daily or weekly basis For example node X could be dedicated to running jobs only from users in the accounting group every Friday from 4 to 10 PM See the Reservation Overview for more information about the use of reservations The Managing Reservations section provides a detailed explanation of the concepts and steps involved in the creation and configuration of standing reservations A standing reservation is a powerful means of Controlling Access to Resources Controlling Turnaround see the following parameters for more information SRNAME SRRESOURCES SRDAYS SRFLAGS SRSTARTTIME SRENDTIME SRWSTARTTIME SRWENDTIME SRDEPTH SRTASKCOUNT SRHOSTLIST SRTPN SRUSERLIST SRGROUPLIST SRACCOUNTLIST SRQOSLIST SRCLASSLIST SRMAXTIME SRTIMELOGIC SRPARTITION SRACCESS Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 7 1 4 Reservation Policies In addition to sta
249. rformed and any problems found are reported Example showstate BOS Summary on Tue May 20 21 18 08 1997 JobName Nodes WCLimit JobState A frl7n11 942 0 16 600 Running B frl5n09 1097 0 32 14100 Starting C fril7n01 942 0 8 6900 Running D fr13n03 24 0 8 28800 Starting E fr15n13 1018 0 8 28800 Starting F frl7n05 953 0 8 86400 Running G fr15n09 1103 0 1 86340 Running H frl3n11 28 0 J 86400 Running I fr14n09 21 0 24 86400 Starting Usage Summary 9 Active Jobs 106 Active Nodes 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 Frame 2 XXXXXXXXXXXXXXXXXXXXXXXX AJ C A C C A Frame 3 A I I Frame 4 I A I E T E Frame 5 F E FI F F I F Frame 6 I I E I I I I F I I T I EE Frame Ts XXX XXX XXX XXX b XXX XXX XXX XXX Frame 9z E Frame T1 I F A I F A Frame 12 A A C A C A A Frame T3 D XXX I XXX XXX XXX XXX XXX I XXX I XXX Frame 14 D XXX I XXX I XXXI D XXX XXX H XXX I XXX XXX Frame L533 b XXX b XXX b XXX b XXX D XXX b XXX b XXX b XXX Frame 16 b XXX XXX b XXX XXX b XXX b XXX XXX b XXX Frame Lah i Frame 21 XXX b XXX b XXX XXX b XXX b XXX b XXX b XXX Frame 22 b XXX b XXX b XXX XXX b XXX Frame 27 b XXX b XXX
250. rity specifies how stale checkpoint CHECKPOINTEXPIRATIONTIME 1 00 00 00 CHECKPOINTEXPIRATIONTIME DD HH MM SS INFINITY data can be before it is ignored Expire checkpoint data which has been stale for over and purged one day CHECKPOINTFILE var adm maui maui ck name absolute or relative of CHECKPOINTFILE lt STRING gt maui ck the Maui checkpoint file Maintain the Maui checkpoint file in the file specified CHECKPOINTINTERVAL 00 15 00 CHECKPOINTINTERVAL DD HH MM SS 00 05 00 ae Dere e oan Ma checkpoints Maui should checkpoint state information every 15 minutes list of zero or more space delimited lt ATTR gt lt VALUE gt pairs where lt ATTR gt is one of the following specifies class specific attributes See the flag overview for a description of CLASSCFG batch MAXJOB 50 QDEF highprio CLASSCFG lt CLASSID gt PRIORITY FSTARGET QLIST QDEF NONE legal flag values up to 50 jobs submitted to the class bat ch will be PLIST PDEF FLAGS or a fairness policy NOTE Only available in Maui allowed to execute simultaneously and will be assigned specification 3 0 7 and higher the QOS highprio by default specifies the weight to be CLASSWEIGHT lt INTEGER gt 0 applied to the class priority of CLASSWEIGHT 10 each job See Cred Factor specifies the
251. rmation Maui does not trust resource manager All node and job information is reloaded on each iteration Discrepancies are logged and handled where possible NodeSyncDeadline JobSyncDeadline overview Purging Stale Information Thread See Also Resource Manager Configuration Resource Manager Extensions Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 13 2 Resource Manager Configuration The type of resource manager to interface to is specified using the RMT YPE parameter This parameter takes an argument of the form lt RMTYPE gt lt RMSUBTYPE gt Currently the following resource manager types are supported LL2 Loadleveler version 2 1 and 2 2 PBS OpenPBS and PBSPro all versions WIKI Text based API used by LRM YRM BProc and other resource managers SGE Sun s Grid Engine Resource Manger The RMSUBTYPE option is currently only used to support Compaq s RMS resource manager in conjunction with PBS In this case the value PBS RMS should be specified As noted above Maui can support more than one resource manager simultaneously Consequently all resource manager parameters are specified as arrays For example to interface to the Loadleveler scheduling API one would specify RMTYPE O LL2 See the Parameters Overview for more information about parameter specification In addition to RMTYPE other parameters allow further control of the scheduler resource manager interface
252. rness throttling policies are considered and scheduled The second pass expands this list of jobs to include those which meet the hard less constrained fairness throttling policies The second important concept regarding Maui backfill is the concept of backfill windows The figure below shows a simple batch environment containing two running jobs and a reservation for a third job The present time is represented by the leftmost end of the box with the future moving to the right The light grey boxes represent currently idle nodes which are eligible for backfill For this example lets assume that the space represented covers 8 nodes and a 2 hour timeframe To determine backfill windows Maui analyzes the idle nodes essentially looking for largest node time rectangles It determines that there are two backfill windows The first window Window 1 consists of 4 nodes which are available for only one hour because some of the nodes are blocked by the reservation for job C The second window contains only one node but has no time limit because this node is not blocked by the reservation for job C It is important to note that these backfill windows overlap Once the backfill windows have been Backfill Windows determined Maui begins to traverse them The current behavior is to traverse these windows widest window first i e most nodes to fewest nodes As each backfill window is evaluated Maui applies the backfill algorithm specif
253. rshare subfactors list of zero or more space delimited specifies group specific GROUPCFG staff MAXJOB 50 lt ATTR gt lt VALUE gt pairs where lt ATTR gt attributes See the flag QDEF highprio is one of the following overview for a description of GROUPCFG lt GROUPID gt PRIORITY FSTARGET QLIST QDEF NONE legal flag values up to 50 jobs submitted by members of the group PLIST PDEF FLAGS or a fairness policy NOTE Only available in Maui S taff will be allowed to execute simultaneously and specification 3 0 7 and higher will be assigned the QOS highprio by default specifies the priority weight GROUPWEIGHT lt INTEGER gt 0 assigned to the specified group GROUPWEIGHT 20 priority See Direct Spec Factor length of time a job is allowed to remain in a starting state If a started job does not transition JOBMAXSTARTT IME 2 00 00 JOBMAXSTARTTIME DD HH MM SS 1 NO LIMIT toa running state within this jobs may attempt to start for up to 2 hours before amount of time the scheduler being cancelled by Maui will cancel the job believing a system failure has occurred amount of time Maui will allow JOBMAXOVERRUN 1 00 00 JOBMAXOVERRUN DD HH MM SS 0 a job to exceed its wallclock allow jobs to exceed their wallclock limit by up to 1 limit before it is terminated hour specifies additional constraints on how compute nodes are to be selected EXACTNODE indicates that Maui sho
254. rvations starting more than 30 enabled in Maui 3 0 7 and minutes in the future higher specifies if a node s load affects its state or its available processors ADJUSTSTATE tells Maui to mark the node busy when MAXLOAD is reached PT EE E E PEAGE Fa causes NODELOADPOLICY ADJUSTSTATE one of the following or node s available procs to be NODELOADPOLICY ADJUSTPROCS E ADJUSTSTATE equivalent to p Maui will mark a node busy if its measured load MIN ConfiguredProcs exceeds its MAXLOAD setting DedicatedProcs MaxLoad CurrentLoad NOTE NODELOADPOLICY only affects a node if MAXLOAD has been set specifies that maximum load on a idle of running node If the NODEMAXLOAD 0 75 NODEMAXLOAD lt DOUBLE gt 0 0 node s load reaches or exceeds Maui will adjust the state of all Idle and Running this value Maui will mark the nodes with a load gt 75 to the state Busy node busy specifies the number of NODEPOLLFREQUENCY 5 NODEPOLLFREQUENCY lt INTEGER gt 0 Poll Always sae acta eae Maui will update node manager based information scheduler initiated node manager queries every 5 scheduling iterations one of FEATURE NETWORK or specifies the type of node attribute by which node set boundaries will be established NODESETATTRIBUTE PROCSPEED
255. s and reservations This command will attempt to immediately start a job Example gt runjob cluster 231 job cluster 231 successfully started This example attempts to run job cluster 231 See Also cancel job cancel a job check job show detailed status of a job showg list queued jobs Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ schedctl Overview The schedctl command controls various aspects of scheduling behavior It is used to manage scheduling activity kill the scheduler and create resource trace files Format schedctl k n I r lt RESUMETIMES gt s S lt ITERATION gt Flags k shutdown the scheduler at the completion of the current scheduling iteration n dump a node table trace to lt STDOUT gt for use in simulations r lt RESUMETIME gt resume scheduling in lt RESUMETIME gt seconds or immediately if not specified s lt ITERATION gt suspend scheduling at iteration lt ITERATION gt or at the completion of the current scheduling iteration if not specified If lt ITERATIONS gt is followed by the letter T maui will not process client requests until this iteration is reached S lt ITERATION gt suspend scheduling in lt ITERATION gt more iterations or in one more iteration if not specified If lt ITERATION gt is followed by the letter T maui will not process client requests until lt ITERATION
256. s may need to be specified depending on the resource manager type Some of the related resource management parameters are listed below Further information about each is available in the parameters documentation RMPORT RMSERVER RMTYPE RMAUTHTYPE RMCONFIGFILE Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 2 3 Initial Maui Testing Maui has been designed with a number of key features that allow testing to occur in a no risk environment These features allow you to safely run Maui in test mode even with your old scheduler running be it an earlier version of Maui or even another scheduler In test mode Maui will collect real time job and node information from your resource managers and will act as if it were scheduling live However its ability to actually affect jobs 1 e start modify cancel etc will be disabled Central to Maui testing is the parameter SERVERMODE This parameter allows administrators to determine how Maui will run The possible values for this parameter are NORMAL TEST and SIMULATION As would be expected to request test mode operation the SERVERMODE parameter must be set to TEST The ultimate goal of testing is to verify proper configuration and operation Particularly the following can be checked e Maui possesses the minimal configuration required to start up e Maui can communicate with the resource manager s e Maui is able to obtain full resource and job
257. s u Let s pretend we need to now take down the entire system for maintenance on Thursday from 2 to 10 PM To do this we would create a reservation gt setres S Let s shutdown the scheduler and call it a day gt schedctl k Using sample traces Collecting traces using Maui Understanding and manipulating workload traces Understanding and manipulating resource traces Running simulation sweeps The stats sim file Is not erased at the start of each simulation run It must be manually cleared or moved if statistics are not to be concatenated Using the profiler tool profiler man page 16 1 Simulation Overview 16 2 Resource Traces 16 3 Workload Traces Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 16 2 Resource Traces Resource traces fully describe all scheduling relevant aspects of a batch system s compute resources In most cases each resource trace describes a single compute node providing information about configured resources node location supported classes and queues etc Each resource trace consists of a single line composed of 21 whitespace delimited fields Each field is described in detail in the table below Field Name nee Data Format Default Value Details Index Resource one of currently the only legal value is Ea RNN COMPUTENODE COMPUTENODE when AVAILABLE DEFINED or DRAINED is one of specified node will start in the AVAILABLE state Idle
258. se the following steps gt SChedce ll s stop Maui scheduling so that the only activity is handling maui client requests gt changeparam LOGLEVEL 7 set the logging level to very verbose gt tail f log maui log more tail the maui log activity In another window gt showq The maui log file should record the client request and any reasons it was rejected If these steps do not reveal the source of the problem the next step may be to check the mailing list archives post a question to the mauiusers list or contact Maui support Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 14 6 Tracking System Failures Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 14 7 Problems with Individual Jobs To determine why a particular job will not start there are several commands which can be helpful checkjob v Checkjob will evaluate the ability of a job to start immediately Tests include resource access node state job constraints ie startdate taskspernode QOS etc Additionally command line flags may be specified to provide further information lt POLICYLEVEL gt evaluate impact of throttling policies on job feasibility n lt NODENAME gt evaluate resource access on specific node r lt RESERVATION_LIST gt evaluate access to specified reservations checknode Display detailed status of node di
259. servation This configuration will tell the scheduler that these nodes can be used but should only be used if it cannot find compute resources elsewhere The final step load balancing is accomplished in two parts First the nodes in group B must be configured to allow up to 4 serial jobs to run at a time This is best accomplished using the PBS virtual nodes feature To load balance simply select the CPULOAD allocation algorithm in Maui This algorithm will instruct Maui to schedule the job based on which node has the most available unused idle CPU time Configuration This site requires both resource manager and scheduler configuration The following Maui configuration would be needed maui cfg reserve overflow processors SRNAME 0 overflow SRHOSTLIST 0 cn0 1 8 hostname regular expression SRCLASSLIST 0 parallel batch use minus sign to indicate negative affinity ALLOCATIONPOLICY CPULOAD allow SMP node sharing NODEACCESSPOLICY SHARED set queue serial resources_max nodeccount 1 set queue serial acl_hosts an0ltan02 anl6 cn0Olt cn02 cn08 set queue serial acl_host_enable true set queue parallel resources_min nodecount 2 set queue parallel acl_hosts bn01 bn02 bn08 cn01 cn02 cn08 set queue parallel acl_host_enable true Monitoring Conclusions Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ A 3 Case Study Development O2K Overview
260. showstats Look at the line Current Active Total Procs to see current system utilization Determine the amount of time associated with each simulated time step gt showconfig grep RMPOLLINTERVAL This value is specified in seconds Thus each time we advance the simulator forward one step we advance the simulation clock forward this many seconds showconfig can be used to see the current value of all configurable parameters Advance the simulator forward one step gt schedctl S schedctl allows you to step forward any number of steps or to step forward to a particular iteration number You can determine what iteration you are currently on using the showstats command s v flag gt showstats v The line statistics for iteration lt X gt specifies the iteration you are currently on You should now be on iteration 2 This means simulation time has now advanced forward lt RMPOLLINTERVALS gt seconds use showg r to verify this change gt showg r Note that the first job will now complete in 4 minutes rather than 5 minutes because we have just advanced now by one minute It is important to note that when the simulated jobs were created both the job s wallclock limit and its actual run time were recorded The wallclock time time is specified by the user indicating his best estimate for an upper bound on how long the job will run The run time is how long
261. ss Allocated Resources With the resources selected Maui then maps job tasks to the actual resources This distribution of tasks is typically based on simple task distribution algorithms such as round robin or max blocking but can also incorporate parallel language library i e MPI PVM etc specific patterns used to minimize interprocesses communication overhead 3 3 2 7 Launch Job With the resources selected and task distribution mapped the scheduler then contacts the resource manager and informs it where and how to launch the job The resource manager then initiates the actual job executable Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 3 4 Configuring the Scheduler Maui is configured using the flat text configfile maui cfg In Maui 3 0 6 and earlier an optional configfile fs cfg could also be specified to define fairshare and QoS configuration In more recent versions this functionality is handled by the CFG parameters within the maui cfg file All config files consist of simple lt PARAMETER gt lt V ALUE gt pairs which are whitespace delimited Parameter names are not case sensitive but lt VALUE gt settings are Some parameters are array values and should be specified as lt PARAMETER gt lt INDEX gt i e QOSCFG hiprio PRIORITY 1000 The lt VALUE gt settings may be integers floats strings or arrays of these Some parameters can be specified as arrays and in suc
262. ss and Constraints 73 2 3 Policy Exemptions 733 Managing QoS Access 7 3 1 QoS Overview The QOS facility allows a site to give special treatment to various classes of jobs users groups etc Each QOS object can be thought of a container of special privileges ranging from fairness policy exemptions to special job prioritization to special functionality access Each QOS object also has an extensive access list of users groups and account which can access these privileges Sites can configure various QOS s each with its own set of priorities policy exemptions and special resource access settings They can then configure user group account and class access to these QOS s A given job will have a default QOS and may have access to several additional QOS s When the job is submitted the submittor may request a specific QOS see the User s Manual for information on specifying job QOS for the resource manager of interest or just allow the default QOS to be used Once a job is submitted a user may adjust the QOS of his job s at any time using the setqos command The setqos command will only allow the user to modify the QOS of his jobs and only change the QOS to a QOS that this user has access to Maui administrators may change the QOS of any job to any value Jobs are currently granted access to a QOS privileges by configuring QDEF QOS Default or QLIST QOS Access List settings in the fs cfg file A job may access a particular QOS if
263. subsystems Object Subsystem Flag Use Account a shows detailed account configuration information detailed account shows detailed account configuration information information Boo detailed fairshare configuration information as FairShare f well as current fairshare usage Frame m shows detailed frame information detailed frame information lt Se e e o detailed shows detailed group information information eoo detailed job information Reports on corrupt job attributes unexpected states and excessive job failures shows detailed node information Reports on Node unexpected node states and resource allocation conditions Partition t shows detailed partition information shows detailed job priority information including priority factor contributions to all idle jobs Queue eee why ineligible jobs or not allowed to run Q shows detailed QOS information akis detailed reservation information Reports on Reservation reservation corruption of unexpected reservation conditions User u shows detailed user information Additionally the checkjob and checknode routines provide detailed information and sanity checking on individual jobs and nodes respectively Using Maui Logs for Troubleshooting Maui logging is extremely useful in determining the cause of a problem Where other systems may be cursed for not providing adequate logging to diagnose a problem Maui may be cursed for the oppos
264. sum of all weighted resources components is then multiplied by the RESWEIGHT parameter and capped by the RESOURCECAP parameter Memory Swap and Disk are all measured in megabytes MB The final resource component PE represents Processor Equivalents This component can be viewed as a processor weighted maximum percentage of total resources factor For example if a job requested 25 of the processors and 50 of the total memory on a 128 processor O2K system it would have a PE value of MAX 25 50 128 or 64 The concept of PE s may be a little awkward to grasp initially but it is a highly effective metric in shared resource systems 5 1 2 4 Service SERV Component The Service component essentially specifies which service metrics are of greatest value to the site Favoring one service subcomponent over another will generally cause that service metric to improve 5 1 2 4 1 QueueTime QUEUETIME Subcomponent In the priority calculation a job s queue time is a duration measured in minutes Use of this subcomponent tends to prioritize jobs in a FIFO order Favoring queue time improves queue time based fairness metrics and is probably the most widely used single job priority metric In fact under the initial default configuration this is the only priority subcomponent enabled within Maui It is important to note that within Maui a job s queue time is not necessarily the amount of time since the job was submitted The parameter JOBPR
265. t AND or OR OR SRRESOURCES 2 PROCS 1 MEM 256 SRTASKCOUNT 2 16 specifies how may tasks should be reserved for the reservation standing reservation 2 will reserve 16 tasks worth of resources in this case 16 procs and 4 GB of real memory specifies how SRMAXTIME access status will be combined with other standing reservation access methods to determine job access If SRTIMELOGIC is set to OR a job is granted access to the reserved resources if it meets the MAXTIME criteria or any other access criteria 1 e SRUSERLIST If SRTIMELOGIC is set to AND a job is granted access to the reserved resources only if it meets the MAXTIME criteria and at least on other access criteria SRMAXTIME 5 1 00 00 SRUSERLIST 5 carol charles SRTIMELOGIC 5 AND Maui will allow jobs from users carol and charles to use up to one hour of resources in standing reservation 5 specifies the minimum number SRTPN 2 4 SRRESOURCES 2 PROCS 2 MEM 256 SRTPN X lt INTEGER gt 0 no TPN constraint of tasks per node which must be Maui must locate at least 4 tasks on each node that is available on eligible nodes to be part of the reservation That is each node included in standing reservation 2 must have at least 8 processors and 1 GB of memory available specif
266. t MAUIHOMEDIR directory when Maui is built Unless specified otherwise Maui will look in this directory for its various config files If you wish to run Maui out of a different directory you can override the default home directory setting by creating a etc maui cfg file containing the string MAUIHOMEDIR lt DIRECTORY gt by setting the environment variable MAUIHOMEDIR or by specifying the configfile explicitly using the C command line option on Maui and the Maui client commands When Maui is run it creates a log file maui log in the log directory and creates a Statistics file in the stats directory with the naming convention stats YYYY_MM_DD e stats 2000_09_20 Additionally a checkpoint file maui ck and lock file maui pid are maintained in the Maui home directory Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 3 2 Scheduling Environment 3 2 1 Scheduling Objects 32 1 1 Jobs oe ee ee ee a S 321 11 Requirement or Req 3 2 1 2 Nodes 3 2 1 3 Advance Reservations 3 2 1 4 Policies 3 2 1 5 Resources 3 2 1 6 Task 3 2 1 7 PE 3 2 1 8 Class or Queue 3 2 1 Scheduling Objects Maui functions by manipulating five primary elementary objects These are jobs nodes reservations QOS structures and policies In addition to these multiple minor elementary objects and composite objects are also utilized These objects are also defined in the scheduling dic
267. t are Running or Starting and consuming CPU resources Displayed are the job name the job s owner and the job state Also displayed are the number of processors allocated to the job the amount of time remaining until the job completes given in HH MM SS notation and the time the job started All active jobs are sorted in Earliest Completion Time First order Idle Jobs are those that are queued and eligible to be scheduled They are all in the Idle job state and do not violate any fairness policies or have any job holds in place The jobs in the Idle section display the same information as the Active Jobs section except that the wall clock CPULIMIT is specified rather than job time REMAINING and job QUEUETIME is displayed rather than job STARTTIME The jobs in this section are ordered by job priority Jobs in this queue are considered eligible for both scheduling and backfilling Non Queued jobs are those that are ineligible to be run or queued Jobs listed here could be in a number of states for the following reasons Idle Job violates a fairness policy Use diagnose q for more information UserHold A LoadLeveler User Hold is in place SystemHold A LoadLeveler System Hold is in place BatchHold Deferred NotQueued A Maui Scheduler Batch Hold is in place used when the job cannot be run because the requested resources are not available in the system or because LoadLeveler has repeatedly failed in attempts to start the job A Maui S
268. t performs a decay weighted average the usage information for that entity contained in the FSDEPTH most recent windows For example assume the entity of interest is user John and the following parameters are set FSINTERVAL 12 00 00 FSDEPTH 4 FSDECAY 0 5 and the fairshare data files contain the following usage amounts for the entity of interest John 0 60 0 Total 0 110 0 John 1 0 0 Total 1 125 0 John 2 10 0 Total 2 100 0 John 3 50 0 Total 3 150 0 The current fairshare usage for user John would calculated as follows Usage 60 541 0 542 10 5 3 50 110 541 125 5 2 100 543 150 Note that the current fairshare usage is relative to the actual resources delivered by the system over the timeframe evaluated not the resources available or configured during that time When configuring fairshare it is important to determine the proper timeframe that should be considered Many sites choose one to two weeks to be the total timeframe covered i e FSDEPTH FSINTERVAL but any reasonable timeframe should work How this timeframe is broken up between the number and length of windows is a matter of preference just note that more windows means that the decay factor will make aged data less significant more quickly Historical fairshare data is organized into a number of data files each file containing the information for a length of time as specified by the FSINTERVAL parameter Although FSDEP
269. t specifed it will return the info for resources accessible to any class Permissions This command can be run by any user Parameters ACCOUNT Account name CLASS Class queue required DURATION Time duration specified as the number of seconds or in DD HH MM SS notation FEATURELIST Colon separated list of node features required GROUP Specify particular group MEMCMP Memory comparison used with the m flag Valid signs are gt gt lt and lt MEMORY Specifies the amount of required real memory configured on the node in MB used with the m flag NODECOUNT Specify number of nodes for inquiry with n flag PARTITION Specify partition to check with p flag QOS Specify QOS to check with q flag USER Specify particular user to check with u flag PARTITION Specify partition to check with p flag Flags A Show backfill information for all users groups and accounts By default showbf uses the default user group and account ID of the user issuing the showbf command a Show backfill information only for specified account d Show backfill information for specified duration g Show backfill information only for specified group h Help for this command m Allows user to specify the memory requirements for the backfill nodes of interest It is important to note that if the optional MEMCMP and MEMORY parameters are used they MUST be enclosed in single ticks to avoid interpretation by the shell For example
270. tes Many of these jobs are interactive in nature Throughout the day large longer running production workload is also submitted but these jobs do not have comparable turnaround time pressure Constraints Must do Nodes in Group A must run only parallel jobs Nodes in Group B must only run serial jobs with up to 4 serial jobs per node Nodes in Group C must not be used unless a job cannot locate resources elsewhere Goals Should do The scheduler should attempt to intelligently load balance the timesharing nodes Analysis As in Case Study 1 The network topology is flat and and nodes are homogeneous within each group The only tricky part of this configuration is the overflow group The easiest configuration is to create two PBS queues serial and parallel with appropriate min and max node counts as desired By default Maui interprets the PBS exclusive hostlist queue attribute as constraining jobs in the queue to run only on the nodes contained in the hostlist We can take advantage of this behavior to assign nodes in Group A and Group C to the queue parallel while the nodes in Group B and Group C are assigned to the queue serial The same can be done with classes if using Loadleveler Maui will incorporate this queue information when making scheduling decisions The next step is to make the scheduler use the overflow nodes of group C only as a last resort This can be accomplished using a negative affinity standing re
271. text reader When profiling statistics stat files covering the time frame of interest should be aggregated into a single file This file can be passed to the profiler command along with a number of configuration flags controlling what data should be processed and how it should be display Command line flags allow specification of constraints such as earliest start date or latest completion date Flags can also be used to evaluate only jobs associated with specific users groups accounts or QOS s Further it is possible to specify that only jobs run on certain nodes be processed Because the trace files are flat text simple UNIX text processing tools such as awk sed or grep can be used to create more elaborate filters should they be needed The output of the profiler command provides extensive detailed information about what jobs ran and what level of scheduling service they received The profiler command documentation should be consulted for more information 9 2 3 FairShare Usage Statistics Regardless of whether of not fairshare is enabled detailed credential based fairshare statistics are maintained Like job traces these statistics are stored in the directory pointed to by the STATDIR parameter Fairshare stats are maintained in a separate statistics file using the format FS lt EPOCHTIMES gt i e FS 982713600 with one file created per fairshare window See the Fairshare Overview for more information These files are also flat text a
272. the remaining 2 processors available for use by any job Stacking reservations is not usually required but some sites choose to do it to enforce elaborate policies There is no problem with doing so so long as you can keep things straight It really is not too difficult a concept just takes a little getting used to See the Reservation Overview section for a more detailed description of reservation use and constraints I am working on extending reservation ACL s to allow cleaner arbitrary ACL list support but there are some significant scheduling performance hits associated with completely general ACL support Now for another example As mentioned earlier by default Maui enforces standing reservations by creating a number of reservations where the number created is controlled by the SRDEPTH parameter When Maui starts up and again each night at midnight Maui updates its periodic non floating standing reservations By default SRDEPTH is set to 2 meaning when Maui starts up it will create two 24 hour reservations covering two days worth of time i e a reservation for today and one for tomorrow At midnight today s reservation will be expired and removed tomorrow s reservation will become today s and Maui will create a new reservation for the next day Maui continues creating reservations in the future as time continues its incessant march forward Everything s great resources are always reserved as needed when today rolls around
273. the scheduler again attempts to schedule it The defer mechanism can be disabled by setting DEFERTIME to 0 To release a job from the defer state issue releasehold a lt JOBID gt The second Maui specific type of hold is known as a batch hold A batch hold is only applied by the scheduler and is only applied after a serious or repeated job failure If a job has been deferred and released DEFERCOUNT times Maui will place it in a batch hold It will remain in this hold until a scheduler admin examines it and takes appropriate action Like the defer state the causes of a batch hold can be determined via checkjob and the hold can be released via releasehold Like most schedulers Maui supports the concept of a job hold Actually Maui supports four distinct types of holds user holds system holds batch holds and defer holds Each of these holds effectively block a job preventing it from running until the hold is removed User Holds User holds are very straightforward Many if not most resource managers provide interfaces by which users can place a hold on their own job which basically tells the scheduler not to run the job while the hold is in place The user may utilize this capability because the job s data is not yet ready or he wants to be present when the job runs so as to monitor results Such user holds are created by and under the control of a non privileged and may be removed at any time by that user As would be expected
274. ther jobs which may not be able to access the reserved resources Normally this is a desired behavior However sometimes it is desirable to reserve resources for use as a last resort ie use the reserved resources only when there are no other resources available This last resort behavior is known as negative affinity Note the dash or negative sign following the special in the SRQOSLIST values above This is not a typo rather it indicates that QOS special should be granted access to this reservation but should be assigned negative affinity Thus the SRQOSLIST parameter specifies that QOS high and low should be granted access with positive affinity use the reservation first where possible and QOS special granted access with negative affinity use the reservation only when no other resources are available Affinity status is granted on a per access object basis rather than a per access list basis and always defaults to positive affinity In addition to negative affinity neutral affinity can also be specified using the character ie SROOSLIST 0 normal high debug low In addition to affinity ACL s may also be of different types Note the SRACCOUNTLIST values in the previous example They are preceded with an exclamation point or NOT symbol This indicates that all jobs with accounts other than projectx and projyectY meet the account ACL Note that ifa lt X gt value ie projectX appears in an ACL line that
275. thin the resource manager internal job and node objects and attributes must be manipulated and placed within Wiki based interface concepts as defined in the Wiki Overview Additionally resource manager parameters must be created to allow a site to configure this interface appropriately Efforts are currently underway to create a new XML based interface with an improved transport and security model This interface will also add support for more flexible resource and workload descriptions as well as resource manager specific policy configuration It is expected that this interface will be available in mid 2002 Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 14 0 Trouble Shooting and System Maintenance 14 1 Internal Diagnostics 14 2 Logging Facilities 14 3 Using the Message Buffer 14 4 Handling Events with the Notification Routine 14 5 Issues with Client Commands ae ae ee a 14 6 Tracking System Failures L 14 7 Problems with Individual Jobs Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 14 1 Internal Diagnostics Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 14 2 Logging Overview The Maui Scheduler provides the ability to produce detailed logging of all of its activities The LOGFILE and or LOGDIR parameters within the maui cfg file specify the destination of this log
276. tion decisions both at the time of starting jobs and at the time of creating reservations The impact of time based issues grows significantly with the number of reservations in place on a given system The LASTAVAILABLE algorithm works on this premise locating resources which Time have the smallest space between the end of a job under consideration and the start of a future reservation Nodes non flat network system On systems where network connections do not resemble a flat all to all topology the placement of tasks may present a significant impact on the performance of communication intensive parallel jobs If latencies and bandwidth of the network between any two nodes vary significantly the node allocation algorithm should attempt to pack tasks of a given job as close to each other as possible to minimize the impact of these bandwidth and latency differences 5 2 2 Resource Based Algorithms Maui contains a number of allocation algorithms which address some of the needs described above Additional homegrown allocation algorithms may also be created and interfaced into the Maui scheduling system The current suite of algorithms is described below 5 2 2 1 CPULOAD Nodes are selected which have the maximum amount of available unused cpu power i e lt of CPU s gt lt CPU load gt Good algorithm for timesharing node systems This algorithm is only applied to jobs starting immediately For the purpose of
277. tionary 3 2 1 1 Jobs Job information is provided to the Maui scheduler from a resource manager such as Loadleveler PBS Wiki or LSF Job attributes include ownership of the job job state amount and type of resources required by the job and a wallclock limit indicating how long the resources are required A job consists of one or more requirements each of which requests a number of resources of a given type For example a job may consist of two requirements the first asking for 1 IBM SP node with at least 512 MB of RAM and the second asking for 24 IBM SP nodes with at least 128 MB of RAM Each requirements consists of one or more tasks where a task is defined as the minimal independent unit of resources By default each task is equivalent to one processor In SMP environments however users may wish to tie one or more processors together with a certain amount of memory and or other resources 3 2 1 1 1 Requirement or Req A job requirement or req consists of a request for a single type of resources Each requirement consists of the following components Task Definition A specification of the elementary resources which compose an individual task Resource Constraints A specification of conditions which must be met in order for resource matching to occur Only resources from nodes which meet all resource constraints may be allocated to the job req Task Count The number of task instances required by the req
278. to the job If identical speed nodes cannot be found the algorithm will allocate the set of nodes with the minimum node speed span or range 5 2 2 8 FASTEST This algorithm will select nodes in fastest node first order Nodes will be selected by node speed if specified If node speed is not specified nodes will be selected by processor speed If neither is specified nodes will be selected in a random order 5 2 2 9 LOCAL This will call the locally created contrib node allocation algorithm See also N A 5 2 3 Time Based Algorithms Under Construction 5 2 4 Locally Defined Algorithms Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 6 0 Managing Fairness Throttling Policies Fairshare and Allocation Management 6 1 Fairness Overview 6 3 Fairshare 2 6 4 Allocation Management Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 6 1 Fairness Overview The concept of fairness varies widely from person to person and site to site To some it implies giving all users equal access to compute resources However more complicated concepts incorporating historical resource usage political issues and job value are equally valid While no scheduler can handle all possible definitions of what fair means Maui provides some flexible tools that help with most common fairness management definitions and needs P
279. to consider scheduling algorithm reservation policies node allocation policies and job prioritization It is almost always a good idea to utilize the scheduler s backfill capability since this has a tendency to increase average system utilization and decrease average turnaround time in a surprisingly fair manner It does tend to favor somewhat small and short jobs over others which is exactly what this site desires Reservation policies are often best left alone unless rare starvation issues arise or quality of service policies are desired Node allocation policies are effectively meaningless since the system is homogeneous The final scheduling aspect job prioritization can play a significant role in meeting site goals To maximize overall system utilization maintaining a significant Resource priority factor will favor large resource processor jobs pushing them to the front of the queue Large jobs though often only a small portion of a site s job count regularly account for the majority of a site s delivered compute cycles To minimize job turnaround the XFactor priority factor will favor short running jobs Finally in order for fairshare to be effective a significant Fairshare priority factor must be included Configuration For this scenario a resource manager configuration consisting of a single global queue class with no constraints would allow Maui the maximum flexibility and opportunities for optimization The following Maui c
280. trained to make certain that only Tom s jobs can use node002 at any time on Friday Advance reservation technology enables many features including backfill deadline based scheduling QOS support and meta scheduling 7 1 1 Reservations Overview 7 1 2 Administrative Reservations 7 1 3 Standing Reservations 7 1 4 Reservation Policies oe a a 7 1 5 Configuring and Managing Reservations Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved 7 1 1 Reservation Overview Every reservation consists of 3 major components a set of resources a timeframe and an access control list Additionally a reservation may also have a number of optional attributes controlling its behavior and interaction with other aspects of scheduling All reservation attributes are described below 7 1 1 1 Resources Under Maui the resources specified for a reservation are specified by way of a task description Conceptually a task can be thought of as an atomic or indivisible collection of resources The resources may include processors memory swap local disk etc For example a single task may consist of one processor 2 GB of memory and 10 GB of local disk A reservation consists of one or more tasks In attempting to locate the resources required for a particular reservation Maui will examine all feasible resources and locate the needed resources in groups specified by the task description An example may help cl
281. uld select as many nodes as requested even if it could pack multiple JOBNODEMATCHPOLICY EXACTNODE JOBNODEMATCHPOLICY zero or more of the following NONE tasks onto the same node In a PBS job with resource specification EXACTNODE or EXACTPROC EXACTPROC indicates that Maui should select only nodes with exactly the number of processors configured as are requested per node even if nodes with excess processors are available nodes lt x gt ppn lt y gt Maui will allocate exactly lt y gt task on each of lt x gt distinct nodes one of the following ALWAYS JOBPRIOACCRUALPOLICY FULLPOLICY QUEUEPOLICY JOBSIZEPOLICY lt N A gt QUEUEPOLICY NONE specifies how the dynamic aspects of a job s priority will be adjusted ALWAYS indicates that the job will accrue queuetime based priority from the time it is submitted FULLPOLICY indicates that it will accrue priority only when it meets all queue AND run policies QUEUEPOLICY indicates that it will accrue priority so long as it satisfies various queue policies i e MAXJOBQUEUED lt N A gt JOBPRIOACCRUALPOLICY QUEUEPOLICY Maui will adjust the job s dynamic priority subcomponents i e QUEUETIME XFACTOR and TARGETQUEUETIME etc each iteration that the job satisfies the associated QUEUE policies such as MAXJOBQUEUED lt N A gt JOBSYNCTIME DD HH MM SS LOGDIR lt STRING gt colon delimited list of one or
282. urable node attributes are listed below NODETYPE The NODETYPE attribute is most commonly used in conjunction with an allocation management system such as QBank In these cases each node is assigned a node type and within the allocation management system each node type is assigned a charge rate For example a site may wish to charge users more for using large memory nodes and may assign a node type of BIGMEM to these nodes The allocation management system would then charge a premium rate for jobs using BIGMEM nodes See the Allocation Manager Overview for more information Node types are specified as simple strings If no node type is explicitly set the node will possess the default node type of IDEFAULT Node type information can be specified directly using NODECFG or through use of the FEATURENODETYPEHEADER parameter Example maui cfg NODECFG node024 NODETYPE BIGMEM PROCSPEED Knowing a node s processor speed can help the scheduler improve intra job efficiencies by allocating nodes of similar speeds together This helps reduce losses due to poor internal job load balancing Maui s Node Set scheduling policies allow a site to control processor speed based allocation behavior Processor speed information is specified in MHz and can be indicated directly using NODECFG or through use of the FEATUREPROCSPEEDHEADER parameter SPEED A node s speed is very similar to its procspeed but is specified as a relative v
283. ver various qualities of service to various users and groups They need to understand how the available resources are being delivered to the various users over time and need the ability to have the administrators tune cycle delivery to satisfy the current site mission objectives How well a scheduler succeeds can only be determined if various metrics are established and a means to measure these metrics are available While statistics are important their value is limited unless optimal statistical values are also known for the current environment including workload resources and policies If one could determine that a site s typical workload obtained an average queue time of 3 hours on a particular system this would be a good statistic However if one knew that through proper tuning the system could deliver an average queue time of 1 2 hours with minimal negative side effects this would be valuable knowledge The Maui Scheduler was developed with extensive feedback from users administrators and managers At its core it is a tool designed to truly manage resources and provide meaningful information about what is actually happening on the system It was created to satisfy real world needs of a batch system administrator as he tries to balance the needs of users staff and managers while trying to maintain his sanity Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved ___ 2 0 Installation Maui inst
284. with approximately the following quartile job frequency distribution 1 2 3 8 9 24 and 25 32 nodes Job Length jobs range in length from 1 to 24 hours Job Owners job are submitted from 6 major groups consisting of a total of about 50 users NOTES During prime time hours the majority of jobs submitted are smaller short running development jobs where users are testing out new code and new data sets The owners of these jobs are often unable to proceed with their work until a job they have submitted completes Many of these jobs are interactive in nature Throughout the day large longer running production workload is also submitted but these jobs do not have comparable turnaround time pressure Constraints Must do The groups Meteorology and Statistics should receive approximately 45 and 35 of the total delivered cycles respectively Nodes cannot be shared amongst tasks from different jobs Goals Should do The system should attempt to minimize turnaround time during primetime hours Mon Fri 8 00 AM to 5 00 PM and maximize system utilization during all other times System maintenance should be efficiently scheduled around Analysis The network topology is flat and and nodes are homogeneous This makes life significantly simpler The focus for this site is controlling distribution of compute cycles without negatively impacting overall system turnaround and utilization Currently the best mechanism for doin
285. x FLAGS ADVRES test job cmd To enable job to reservation mapping via QoS the QoS flag USERRESERVED should be set in a similar manner 7 1 1 5 Reservation Specification There are two main types of reservations which sites typically deal with The first administrative reservations are typically one time reservations created for special purposes and projects These reservations are created using the setres command These reservations provide an integrated mechanism to allow graceful management of unexpected system maintenance temporary projects and time critical demonstrations This command allows an administrator to select a particular set of resources or just specify the quantity of resources needed For example an administrator could use a regular expression to request a reservation be created on the nodes blueO 1 9 or could simply request that the reservation locate the needed resources by specifying a quantity based request such as TASKS 20 The second type of reservation is called a standing reservation It is of use when there is a recurring need for a particular type of resource distribution For example a site could use a standing reservation to reserve a subset of its compute resources for quick turnaround jobs during business hours on Monday thru Friday Standing reservations are created and configured by specifying parameters in the maui cfg file The Standing Reservation Overview provides more information about configuri
286. y The NOTIFICATIONPROGRAM parameter allows a site to specify the name of the program to run This program is most often locally developed and designed to take action based on the event which has occurred The location of the notification program may be specified as a relative or absolute path If a relative path is specified Maui will look for the notification relative to the MAUIHOMEDIR tools directory In all cases Maui will verify the existence of the notification program at start up and will disable it if it cannot be found or is not executable The notification program s action may include steps such as reporting the event via email adjusting scheduling parameters rebooting a node or even recycling the scheduler For most events the notification program is called with commandline arguments in a simple lt EVENTTYPE gt lt MESSAGE gt format The following event types are currently enabled Event Type Event Type Format ss Format _ Description Maui cannot BANKFAILURE lt MESSAGE gt successfully JOBCORRUPTION lt MESSAGE gt communicate with the bank due to reasons such as connection failures bank corruption or parsing failures An active job is in an unexpected state or has one or more allocated nodes which are in unexpected states JOBHOLD lt MESSAGE gt A job hold has been placed on a job JOBWCVIOLATION lt MESSAGE gt A job has exceeded its wallclock limit Reservatio
287. y available memory limits REQUEUE or SUSPEND in Maui 3 2 and higher where RESOURCE is one or more of PROC DISK SWAP or MEM RESWEIGHT 0 5 MEMORYWEIGHT 0 10 all resource priority components pROCWEIGHT 0 100 RESWEIGHT X lt INTEGER gt 0 are miltipli d Oy this value SWAPWEIGHT 0 0 before being added to the total RESOURCECAP 0 2000 job priority the job priority resource factor will be calculated as MIN 2000 5 10 JobMemory 100 JobProc one of CHECKSUM PKL or specifies the security protocol to RMAUTHTYPE 0 CHECKSUM RMAUTHTYPE X SECUREPORT f CHECKSUM be used in scheduler resource The scheduler will require a secure checksum manager communication associated with each resource manager message RMNAME 0 DevCluster specifies name of resource RMNAME X lt STRING gt lt X gt manager lt X gt resource manager 0 will be referred to as DevCluster in maui command output and maui logs specifies a non default RM node RMNMPORT 0 13001 manager through which RMNMPORTIX lt INTEGER gt any valid port number extended node attribute Maui will contact the node manager located on each information may be obtained compute node at port 13001 RMPOLLINTERVAL 60 RMPOLLINTERVAL DD HH MM SS 00 01 00 specifies interval between RM Maui will refresh its resource manager information polls every 60 seconds NOTE this parameter specifies the global poll interval for all resource managers specifies the port on which
288. y to use or override most major Maui functions The purpose of this library is to allow the development and use of extension modules or plug ins similar to those available for web browsers One such library G2 currently extends many core Maui capabilities in the areas of resource management resource allocation and scheduling Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ Appendix C Adding New Algorithms with the Local Interface Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved Appendix D Adjusting Defaulting Limits Maui is distributed in a configuration capable of supporting multiple architectures and systems ranging from a few processors to several thousand processors However in spite of its flexibility it still contains a number of archaic static structures defined in header files These structures limit the default number of jobs reservations nodes etc which Maui can handle and are set to values which provide a reasonable compromise between capability and memory consumption for most sites However many sites desire to increase some of these settings to extend functionality or decrease them to save consumed memory The most common parameters are listed below and can be adjusted by simply modifying the appropriate define and rebuilding Maui Parameter Location Default Max Description Tested total numbe
289. you are simply trying to become familiar with the flow of the scheduler The scheduler can be run with a low LOGLEVEL value at first to show the highest level functions This shows high level data and control flow Increasing the LOGLEVEL increases the number of functions displayed as familiarity with the scheduler flow grows The LOGLEVEL can be changed on the fly by use of the changeparam command or by modifying the maui cfg file and sending the scheduler process a SIGHUP Also if the scheduler appears to be hung or is not properly responding the LOGLEVEL can be incremented by one by sending a SIGUSR1 signal to the scheduler process Repeated SIGUSRI signals will continue to increase the LOGLEVEL The SIGUSR2 signal can be used to decrement the LOGLEVEL by one If an unexpected problem does occur save the log file as it is often very helpful in isolating and correcting the problem Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 14 3 Using the Message Buffer Under Construction Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ 14 4 Handling Events with the Notification Routine Maui possesses a primitive event management system through the use of the notify program The program is called each time an event of interest occurs Currently most events are associated with failures of some sort but use of this facility need not be limited in this wa
290. yright 2000 2002 Supercluster Research and Development Group All Rights Reserved setqos setgos h QOS JOB Purpose Set Quality Of Service for a specified job Permissions This command can be run by any user Parameters JOB Job number QOS Quality Of Service level Range is 0 lowest to 8 highest Jobs default to a QOS level of 0 unless the user group or account has a different value specified in the fairshare configuration file s cfg Users are allowed to set the QOS for their own jobs in the range of 0 to the maximum value allowed by the user group and or account which owns the job Flags h Help for this command Description This command allows you to set the Quality Of Service QOS level for a specified job Users are allowed to use this command to change the QOS of their own jobs Example setgos 3 fr28n13 1198 0 Job QOS Adjusted This example sets the Quality Of Service to a value of 3 for job number fr28n13 1198 0 Related Commands None Default File Location u loadl maui bin setgos Notes None Copyright 1998 Maui High Performance Computing Center All rights reserved Copyright 2000 2002 Supercluster Research and Development Group All Rights Reserved __ setres setres ARGUMENTS lt RESOURCE_EXPRESSION gt ARGUMENTS a lt ACCOUNT_LIST gt c lt CHARGE SPEC gt d lt DURATION gt e lt ENDTIME gt f lt FEATURE_LIST gt g

Download Pdf Manuals

Related Search

Related Contents

Release Notes  Vêtements, apparences et travail social  AUFBAU-UND BEDIENUNGSANLEITUNG  Plug-Ins Manager 5 User Manual  A school kid runs by. I tell him it`s electric. He says he knows  Manuel de l`inspecteur Navigabilité Chapitre xxx  CEMbox 750 l  PMLLPADCAN-L: Manuel du produit  The PM25 Controller product is supplied, designed and  Philips SHH1110 Headphones to phone connector  

Copyright © All rights reserved. Failed to retrieve file