Home

BXOFED User‟s Manual

1. Partition your hard Expert Partitioner disks This is intended for experts If you are not Device Size Type Mount Mount By Start End Used By Label Devi Tamiliarw hie Jdev sda 8 0GB IETVIRTUALDISK 0 1045 scsi conceptsuthard disk Jdev sdal 70 5 8 Linux native Ext2 boot 0 8 scsi partitiongend haw c dev sda2 502 0 MB F Linux swap swap use them you might z wantto go back and Idev sda3 F Linux native Reiser 73 1045 select automatic partitioning Please note that nothing will be written to your hard disk until you confirm Really delete device dev sda2 the entire installation in the last installation dialog Until that point you can safely abort the installation For LVM setup using a non LVM root device and a non LVM swap device is recommended Other than the root and swap devices you should have partitions managed by LVM The table to the right shows the current partitions on all your Create Edit Delete Resize J hard disks MENS LVM EVMS RAID v Crypt File Expert v Step 13 In the pop up window click No to approve deleting the swap partition You will be returned to Installation Settings window See image below Mellanox Technologies Rev 1 10 190 Partition your hard Expert Partitioner disks This is intended for exper
2. 92 perfmgr sweep time s seconds Specify th Sweep time for the performance manager in sec onds default is 180 seconds Only takes effect if enable perfmgr was specified at configure time consolidate ipv6 snm req Consolidate IPv6 Solicited Node Multicast group join requests into one multicast group per MGID PKey v verbose This option increases the log verbosity level The v option may be specified multiple times to further increase the ver bosity level See the D option for more information about log verbosity V This option sets the maximum verbosity level and forces log flushing The V option is equivalent to D OxFF d 2 See the D option for more information about log verbosity D This option sets the log verbosity level A flags field must follow the D option A bit set clear in the flags enables disables a specific log level as follows BIT OG LEVEL ENABLED 0x01 ERROR error messages 0x02 INFO basic messages low volume 0x04 VERBOSE interesting stuff moderate volume 0x08 DEBUG diagnostic high volume 0x10 FUNCS function entry exit very high volume 0x20 FRAMES dumps all SMP and GMP frames 0x40 ROUTING dump FDB routing information 0x80 currently unused Without D OpenSM defaults to ERROR INFO 0x3 Specify ing D 0 disables all messages Specifying D OxFF enables all messages see V High verbosity levels may require increasing the transactio
3. 1X or 4X LinkWidthA ctive 4X LinkSpeedSupported 2 5 Gbps LinkSpeedEnabled 2 5 Gbps LinkSpeedActive 2 5 Gbps Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 130 3 Change the speed of a port First query for current configuration gt ibportstate C mlx4 0 D 0 1 PortInfo Port info DR path slid 65535 dlid 65535 0 port 1 eee ee Initialize PhysLinkState LinkUp LinkWidthSupported 1X or 4X LinkWidthEnabled 1X or 4X LinkWidthA ctive 4X LinkSpeedSupported 2 5 Gbps or 5 0 Gbps LinkSpeedEnabled 2 5 Gbps or 5 0 Gbps LinkSpeedA ctive 5 0 Gbps Now change the enabled link speed gt ibportstate C mlx4 0 D 0 1 speed 2 ibportstate C mlx4 0 D 0 1 speed 2 Initial PortInfo Port info DR path slid 65535 dlid 65535 0 port 1 LinkSpeedEnabled 2 5 Gbps After PortInfo set Port info DR path slid 65535 dlid 65535 0 port 1 LinkSpeedEnabled 5 0 Gbps IBA extension Show the new configuration gt ibportstate C mlx4_0 D 0 1 PortInfo Port info DR path slid 65535 dlid 65535 0 port 1 Einkstate ee s Initialize PhysLinkState LinkUp LinkWidthSupported or 4X Link WidthEnabled 1X or 4X
4. A S 129840354 Revbata 129529906 AN 1803332 RcvBkts mn 1799018 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 141 3 Read then reset performance counters from LID 2 port 1 gt perfquery r 21 Port counters Lid 2 port 1 11 12 ibcheckerrs Applicable Hardware All InfiniBand devices Description Validates an IB port or node and reports errors in counters above threshold Check specified port or node and report errors that surpassed their predefined threshold Port address is lid unless G option is used to specify a GUID address The predefined thresholds can be dumped using the s option and a user defined threshold file using the same format as the dump can be specified using the t lt file gt option Synopsis ibcheckerrs h b v G T threshold file gt s N nocolor C ca name P ca port t timeout ms lt lid guid gt lt port gt Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 142 Table 32 lists the various flags of the command Table 32 ibcheckerrs Flags and Options Fla Optional Default Description Mandatory If Not Specified P h help Optional Print the help menu b Optional Print in brief mode Reduce the output to show only if errors are present not what they are v erbose Optional Increase verbosity level May be
5. The DHCP server must run on a machine which has loaded the IPoIB module To run the DHCP server from the command line enter dhcpd IB network interface name gt d Example 1 dhcpd ib0 d DHCP Client Optional Note A DHCP client can be used if you need to prepare a diskless machine with an IB driver See Step 8 under Example Adding an IB Driver to initrd Linux Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 31 In order to use a DHCP client identifier you need to first create a configuration file that defines the DHCP client identifier Then run the DHCP client with this file using the following command dhclient cf client conf file IB network interface name Example of a configuration file for the ConnectX PCI Device ID 25418 called dhc1ient conf The value indicates a hexadecimal number interface ibl send dhcp client identifier 00 02 c9 03 00 00 10 39 Example of a configuration file for InfiniHost III Ex PCI Device ID 25218 called dhclient conf The value indicates a hexadecimal number interface ibl send dhcp client identifier 20 00 55 04 01 fe 80 00 00 00 00 00 00 00 02 c9 02 00 23 13 92 In order to use the configuration file run host1 dhclient cf dhclient conf ibl 3 3 2 Static IPoIB Configuration If you wish to use an IPoIB configuration that is not based on DHCP you need to supply the instal lat
6. cd usr local include scst cp scst h scsi tgt h patch p0 lt tmp scsi tgt patch Note The above steps are for RHEL 5 2 distributions only e Savethe following patch as tmp scst patch scst h2008 07 20 14 25 30 000000000 0700 scst h2008 07 20 14 25 09 000000000 0700 42 7 42 9 endif if LINUX VERSION CODE lt KERNEL VERSION 2 6 19 typedef Bool bool define true 1 define false 0 endif f Untar patch and install scst 1 0 0 tar zxvf scst 1 0 0 tar gz cd scst 1 0 0 include patch p0 tmp scst patch cd make amp amp make install g Savethe following patch as tmp scsi tgt patch scsi_tgt h2008 07 20 14 25 30 000000000 0700 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 203 scsi_tgt h2008 07 20 14 25 09 000000000 0700 2330 7 2332 7 void scst mcmd completed struct sest mgmt cmd mcmd int status if LINUX VERSION CODE lt KERNEL VERSION 2 6 24 static inline struct page sg page struct scatterlist sg 1 return sg gt page 2358 7 2360 7 sg gt offset offset sg gt length len endif LINUX VERSION CODE lt KERNEL VERSIONQ 6 24 static inline void sg clear struct scatterlist sg h Patch tgt h with tmp scsi_tgt patch cd usr local include scst cp scst h scsi tgt h patch p0 lt tmp scsi tgt patch 4 When you ins
7. Mellanox BXOFED Stack for Linux User s Manual 185 Step 5 The iSCSI Initiator Discovery window will now request authentication to access the iSCSI target Click Next to continue without authentication unless authentication is required Preparation Language License Agreement gt Disk Activation e System Analysis e Time Zone 3 iSCSI Initiator Discovery Installation e Installation Summary e Perform Installation Configuration e Root Password Hostname Network Customer Center Incoming Authentication e Online Update Username Password e Service X No Authentication Users Clean Up Release Notes Hardware Configuration Outgoing Authentication Username Password Help Back Abort Step 6 The iSCSI Initiator Discovery window will show the iSCSI target that got connected to Note that the Connected column must indicate True for this target Click Next See figure below Mellanox Technologies Rev 1 10 186 Preparation Language License Agreement gt Disk Activation e System Analysis Connected e Time Zone 10 4 3 3 oot True iSCSI Initiator Discovery Installation e Installation Summary e Perform Installation Configuration e Root Password Hostname e Network e Customer Center e Online Update e Service e Users e Clean Up e Release Notes e Hardware Configuration Ste
8. new routing modules Cur rently only unicast callbacks are supported Multicast can be added later One existing routing module is up down updn which may be activated with R updn option instead of old u General usage is hostl opensm R module name There is also a trivial routing module which is able to load LFT tables from a dump file Main features are This will load switch LFTs and or LID matrices min hops tables This will load switch LFTs according to the path entries introduced in the dump file e No additional checks will be performed such as is port connected etc n case when fabric LIDs were changed this will try to reconstruct LFTs correctly if endport GUIDs are represented in the dump file in order to disable this GUIDs may be removed from the dump file or zeroed The dump file format is compatible with output of ibroute utility and for whole fabric can be generated with dump _Ifts sh script To activate file based routing module use hostl opensm R file U path to dump file If the dump file is not found or is in error the default routing algorithm is utilized The ability to dump switch lid matrices aka min hops tables to file and later to load these is also supported The usage is similar to unicast forwarding tables loading from dump file introduced by file routing engine but new lid matrix file name should be specified by M or 1id matrix file option For examp
9. role lt program name gt lt address gt lt port range gt where lt address family gt Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 57 can be one of Sdp for specifying when an SDP should be used tcp for specifying when an SDP socket should not be matched both for specifying when both SDP and AF INET sockets should be used Note that both semantics is different for server and client roles For server it means that the server will be listening on both SDP and TCP Sockets For client the connect function will first attempt to use SDP and will silently fall back to TCP if the SDP connection fails lt role gt can be one of server or listen for defining the listening port address family client or connect for defining the connected port address family lt program name gt Defines the program name the rule applies to not including the path Wildcards with same semantics as ls are supported and 2 So db2 would match on any program with a name starting with db2 t cp would match on ttcp etc If program name is not provided default the statement matches all programs lt address gt Either the local address to which the server binds or the remote server address to which the client connects The syntax for address matching is lt IPv4 address gt lt prefix length gt IPv4 address 0 9 0 9 0 9 0 9 each
10. 10 unnamed system 00 00 00 00 00 05 67 90 up Other useful options for this script include h y e print usage print script version list all virtual nics print short info 47 Mellanox Technologies Rev 1 10 EoIB 48 4 5 2 ethtool Another way to retrieve interface info and change its configuration 1s through the use of the stan dard ethtool application EoIB interfaces support ethtool in a similar way to HW Ethernet inter faces The supported Ethtool options include the following options Eon C Show and update interrupt coalesce options g Query RX TX ring parameters k K Show and update protocol offloads i Show driver information S Show adapter statistics For more information on ethtool run ethtool h 4 5 3 Link state An EoIB interface has two different link states that it can report The first is the physical link state of the interface and the second is the link state of the external port associated with the vNic inter face The physical link state of the port is by itself made up of the actual HCA port link state and the status of the vNics connection with the BridgeX If the HCA port link state is down or the EoIB connection with the BridgeX has failed the link will be reported as down This is because without the connection to the BridgeX the EoIB protocol does not work so no data can be sent on the wire The mlx4 vnic driver can also report the status of the external BridgeX port
11. Hostname Root Password Network Customer Center Online Update All information required forthe base installation is now complete e Service If you continue now partitions on your hard disk will be formatted erasing any existing data in those partitions according to the installation settings in the previous dialogs Go back and check the settings if you are unsure Users Clean Up Release Notes Hardware Config Show Release Change v Step 19 At the end of the file copying stage the Finishing Basic Installation window will pop up and ask for confirming a reboot You can click OK to skip count down See image below Note Assuming that the machine has been correctly configured to boot from BoIB via its con nection to the iSCSI target make sure that MLNX IB for ConnectX family or gPXE for InfiniHost III family has the highest priority in the BIOS boot sequence Mellanox Technologies Rev 1 10 171 172 Preparation v Language v License Agreement v System Analysis we Time Zone Copy files to installed system Installation v Installation Summary Save configuration gt Perform Installation Install boot manager Configuration Save installation settings e Hostname e Root Password Prepare system for initial boot e Network e Customer Center e Online Update e Service Users e Clean Up e Release Notes The system will reboot now
12. LinkWidthA ctive 4X LinkSpeedSupported 2 5 Gbps or 5 0 Gbps LinkSpeedEnabled 5 0 Gbps IBA extension LinkSpeedA ctive 5 0 Gbps Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 11 9 ibroute Applicable Hardware InfiniBand switches Description Uses SMPs to display the forwarding tables unicast LinearForwardingTable or LFT or multicast MulticastForwardingTable or MFT for the specified switch LID and the optional lid mlid range The default range is all valid entries in the range 1 to FDBTop Synopsis ibroute h d v V a n D G M s lt smlid gt C lt ca_name gt P lt ca_port gt t lt timeout_ms gt lt dest dr path lid guid lt startlid gt lt endlid gt Table 29 lists the various flags of the command Table 29 ibportstate Flags and Options 131 Fla Optional Default Description 8 Mandatory If Not Specified h help Optional Print the help menu d ebug Optional Raise the IB debug level May be used several times for higher debug levels ddd or d d d a ll Optional Show all LIDs in range including invalid entries v erbose Optional Increase verbosity level May be used several times for additional verbosity vvv or v v v V ersion Optional Show version info a ll Optional Show all LI
13. OpenSM QoS policy file Note In the following policy file example replace OST and MDS with the real port GUIDs qos ulps default 0 default SL for any target port guid OST1 0ST2 OST3 OST4 1 SL for Lustre OST any target port guid MDS1 MDS2 2 SL for Lustre MDS Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 115 end qos ulps OpenSM options file qos max vis 8 qos high limit 0 qos vlarb high 2 1 qos vlarb low 0 96 1 224 qoOs sli2yl 051 2 3 4 5 0 0 15 T 5 15 15 15 175 15 15 10 7 2 EDC SOA 2 tier IPoIB and SRP The following is an example of QoS configuration for a typical enterprise data center EDC with service oriented architecture SOA with IPoIB carrying all application traffic and SRP used for storage QoS Levels Application traffic IPoIB UD and CM and SDP Isolated from storage Min BW of 5096 SRP Min BW 5095 Bottleneck at storage nodes Administration OpenSM QoS policy file Note In the following policy file example replace SRPT with the real SRP Target port GUIDs qos ulps default ipoib sdp h v n srp target port guid SRPT1 SRPT2 SRPT3 end qos ulps OpenSM options file qos max vls 8 qos high limit 0 qos vlarb high 1 32 2 32 qos vlarb low 0 1 qos sI2vl 0 1 2 3 4 5 0 7 15 15 15 15 15 15 15 15 Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 116 10 7 3 EDC 3 tier IPoIB
14. The algorithm also dumps compute node ordering file opensm ftree ca order dump in the same directory where the OpenSM log resides This ordering file provides the CN order that may be used to create efficient communication pattern that will match the routing tables Activation through OpenSM Use R ftree option to activate the fat tree algorithm e Use a lt root_guid file gt to provide root nodes for ranking If the a option is not used rout ing algorithm will detect roots automatically e Use u root cn file gt to provide the list of compute nodes If the u option is not used all the CAs are considered as compute nodes Note LMC gt 0 is not supported by fat tree routing If this is specified the default routing algo rithm is invoked instead 10 5 5 LASH Routing Algorithm LASH is an acronym for LAyered SHortest Path Routing It is a deterministic shortest path routing algorithm that enables topology agnostic deadlock free routing within communication networks When computing the routing function LASH analyzes the network topology for the shortest path routes between all pairs of sources destinations and groups these paths into virtual layers in such a way as to avoid deadlock Note LASH analyzes routes and ensures deadlock freedom between switch pairs The link from HCA between and switch does not need virtual layers as deadlock will not arise between switch and HCA In more deta
15. The table to the right shows the current partitions on all your Create Edit Delete Resize J hard disks MENS LVM EVMS RAID v Crypt File Expert v Step 13 In the pop up window click No to approve deleting the swap partition You will be returned to Installation Settings window See image below Mellanox Technologies Rev 1 10 168 Partition your hard disks This is intended for experts If you are not familiar with the concepts of hard disk partitions and how to use them you might wantto go back and select automatic partitioning Please note that nothing will be written to your hard disk until you confirm the entire installation in the last installation dialog Until that point you can safely abort the installation For LVM setup using a non LVM root device and a non LVM swap device is recommended Other than the root and swap devices you should have partitions managed by LVM The table to the right showsthe current partitions on all your hard disks B Hard disks are Expert Partitioner Size Type Mount By Start End Used By Label Devi 8 0 GB 70 5 MB 7 4 GB IET VIRTUAL DISK 0 1045 F Linux native Ext2 boo K 0 8 F Linux native Reiser 73 1045 You have not assigned a swap partition There is nothing wrong with that but in most cases i
16. e Hardware Configuration 8 Finished Step 20 Once the boot is complete the Startup Options window will pop up Select SUSE Linux Enterprise Server 10 SP2 then press Enter SUSE Linux Enterprise Server 10 Floppu SUSE Linux Enterprise Server 10 Failsafe Boot Options Step 21 The Hostname and Domain Name window will pop up Continue configuring your machine until the operating system is up then you can start running the machine in normal operation mode Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 173 Step 22 Optional If you wish to have the second instance of connecting to the iSCSI Target go through the IB driver copy the initrd file under boot to a new location add the IB driver into it after the load commands of the iSCSI Initiator modules and continue as described in Section A 8 on page 156 Warning Pay extra care when changing initrd as any mistake may prevent the client machine from booting It is recommended to have a back up iSCSI Initiator on a machine other than the client you are working with to allow for debug in case initrd gets corrupted In addition edit the init file that is in the initrd zip and look for the following string if iSCSI TARGET IPADDR then iscsiserver 1SCSI TARGET IPADDR fi Now add before the string the following line iSCSI TARGET IPADDR IB IP Address of iSCSI Target Example iSCSI TARGET IPADDR 11 4 3
17. etc init d openibd restart Mellanox Technologies Rev 1 10 RDS 52 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 6 6 1 6 2 6 3 SDP Overview Sockets Direct Protocol SDP is an InfiniBand byte stream transport protocol that provides TCP stream semantics Capable of utilizing InfiniBand s advanced protocol offload capabilities SDP can provide lower latency higher bandwidth and lower CPU utilization than IPoIB running some sockets based applications SDP can be used by applications and improve their performance transparently that is without any recompilation Since SDP has the same socket semantics as TCP an existing application is able to run using SDP the difference is that the application s TCP socket gets replaced with an SDP socket It is also possible to configure the driver to automatically translate TCP to SDP based on the source IP port the destination or the application name See Section 6 5 The SDP protocol is composed of a kernel module that implements the SDP as a new address fam ily protocol family and a library see Section 6 2 that is used for replacing the TCP address family with SDP according to a policy This chapter includes the following sections e ibsdp so Library on page 53 Configuring SDP page 53 Environment Variables page 56 Converting Socket based Applications page 56 Note that the default valu
18. ompi ver tests osu benchmarks osu ver osu latency OSU MPI Latency Test v3 0 H Size Latency us 0 1 23 1 1 37 2 1 55 4 1 54 8 1 55 16 1 58 32 1 59 64 1 59 128 1 78 256 2 05 512 2 69 1024 3 75 2048 6 14 4096 10 07 Mellanox Technologies Rev 1 10 MPI 80 8192 12 77 16384 18 36 32768 30 52 65536 48 92 131072 93 18 262144 171 92 524288 341 08 1048576 737 97 2097152 1729 27 4194304 4226 58 8 6 4 Intel MPI Benchmark To run the Intel MPI Benchmark test enter host1 usr mpi gcc openmpi ompi ver bin mpirun np 2 mca mpi leave pinned 1 hostfile home lt username gt cluster usr mpi gcc openmpi lt ompi ver gt tests IMB lt IMB ver gt IMB MPI1 Date Mon Mar 10 12 57 18 2008 Machine x86 64 System Linux Release 2 6 16 21 0 8 smp Version 1 SMP Mon Jul 3 18 25 39 UTC 2006 MPI Version 2 0 MPI Thread Environment MPI THREAD SINGLE Minimum message length in bytes 0 Maximum message length in bytes 4194304 MPI Datatype MPI BYTE MPI Datatype for reductions MPI FLOAT MPI Op MPI SUM ot dk H H List of Benchmarks to run PingPong PingPing Sendrecv Exchange Allreduce Reduce Reduce_scatter Allgather Allgatherv Alltoall Alltoallv Beast Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual Barrier Benchmarking PingPong processes 2 bytes repetition
19. All other end ports are assigned partial membership 10 4 1 File Format Notes Line content followed after character is comment and ignored by parser General File Format Partition Definition PortGUIDs list gt Partition Definition PartitionName PKey flag value defmember full limited where PartitionName string will be used with logging When omitted an empty string will be used PKey P Key value for this partition Only low 15 bits will be used When omitted P Key will be autogenerated flag used to indicate IPoIB capability of this partition defmember full limited Specifies default membership for port guid list Default is limited Currently recognized flags are ipoib indicates that this partition may be used for IPOIB asa result IPoIB capable MC group will be created rate val specifies rate for this IPoIB MC group default is 3 10GBps mtu val specifies MTU for this IPoIB MC group default is 4 2048 sl lt val gt specifies SL for this IPoIB MC group default is 0 scope lt val gt specifies scope for this IPoIB MC group default is 2 link local Note that values for rate mtu and scope should be specified as defined in the IBTA specifica tion for example mtu 4 for 2048 Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 98 PortGUIDS list PortGUID GUID of partition member EndPort Hexadecimal numbers should start from
20. B 4 TFTP Server If you have set the filename parameter in your DHCP configuration to a non empty filename you need to install TFTP Trivial File Transfer Protocol TFTP is a simple FTP like file transfer proto col used to transfer files from the TFTP server to the boot client as part of the boot process B 5 BIOS Configuration The expansion ROM image presents itself to the BIOS as a boot device As a result the BIOS will add MLNX NIC lt ver gt to the list of boot devices The priority of this list can be modified through BIOS setup B 6 Operation B 6 1 Prerequisites Make sure that your client is connected to the server s The ConnectX EN PXE image is already programmed on the adapter card see Section B 2 Configure and start the DHCP server as described in Section B 3 Configure and start at least one of the services iSCSI Target see Section B 8 and or TFTP see Section B 4 B 6 2 Starting Boot Boot the client machine and enter BIOS setup to configure MLNX NIC ver to be the first on the boot device priority list see Section B 5 Note On dual port network adapters the client first attempts to boot from Port 1 If this fails it switches to boot from Port 2 If MLNX NIC was selected through BIOS setup the client will boot from ConnectX EN PXE The client will display ConnectX EN PXE attributes and will attempt to bring up a port link Mellanox ConnectX gPXE 0 9 5 Open Source Boot Firmware metO 00 02 c
21. E int ip ret inet_aton argv 1 amp to addr sin addr if ip ret 0 printf invalid ip address s n argv 1 exit EXIT FAILURE j int conn ret connect sd struct sockaddr amp to addr sizeof to addr if conn ret 0 perror connect failed exit EXIT FAILURE printf connected to s u n inet_ntoa to_addr sin_addr ntohs to_addr sin_port 59 Mellanox Technologies Rev 1 10 SDP 60 ssize t nw write sd tx buffer TXBUFSZ if nw lt 0 perror write failed exit EXIT FAILURE else if nw 0 printf socket was closed by remote host n printf sent zd bytes n nw close sd return 0 sdp_server c Code Usage sdp server include lt stdio h gt include lt stdlib h gt include lt stdint h gt include lt unistd h gt include lt sys types h gt include lt sys socket h gt include lt netinet in h gt include lt arpa inet h gt include lt sys epoll h gt include lt errno h gt include lt assert h gt define RXBUFSZ 2048 uint8 trx buffer RXBUFSZ define DEF PORT 22222 define AF INET SDP 27 define PF INET SDP AF INET SDP int main int argc char argv int sd socket PF INET SDP SOCK STREAM 0 if sd lt 0 perror socket failed exit EXIT FAILURE Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual struc
22. Fibre Channel over Ethernet 1 4 5 ULPs IPoIB The IP over IB IPoIB driver is a network interface implementation over InfiniBand IPoIB encap sulates IP datagrams over an InfiniBand connected or datagram transport service IPoIB pre appends the IP datagrams with an encapsulation header and sends the outcome over the InfiniBand transport service The transport service is Reliable Connected RC by default but it may also be configured to be Unreliable Datagram UD The interface supports unicast multicast and broad cast For details see Chapter 3 IPoIB EoIB The Ethernet over IB EoIB mlx4 vnic module is a network interface implementation over InfiniBand EoIB encapsulates Layer 2 datagrams over an InfiniBand Datagram UD transport service The InfiniBand UD datagrams encapsulates the entire Ethernet L2 datagram and its payload For details see Chapter 4 EoIB RDS Reliable Datagram Sockets RDS is a socket API that provides reliable in order datagram delivery between sockets over RC or TCP IP For more details see Chapter 5 RDS SDP Sockets Direct Protocol SDP is a byte stream transport protocol that provides TCP stream seman tics SDP utilizes InfiniBand s advanced protocol offload capabilities Because of this SDP can have lower CPU and memory bandwidth utilization when compared to conventional implementa tions of TCP while preserving the TCP APIs and semantics upon which most current netw
23. For the transparent integration with SDP the following two environment variables are required 1 LD PRELOAD this environment variable is used to preload 1ibsdp so and it should point to the 1ibsdp so library The variable should be set by the system administrator to usr 1ib libsdp so or usr lib64 libspd so 2 LIBSDP CONFIG FILE this environment variable is used to configure the policy for replac ing TCP sockets with SDP sockets By default it points to etc 1ibsdp conf SIMPLE LIBSDP ignore 1ibsdp conf and always use SDP 6 5 Converting Socket based Applications You can convert a socket based application to use SDP instead of TCP in an automatic also called transparent mode or in an explicit also called non transparent mode Automatic Transparent Conversion The 1ibsdp conf configuration policy file is used to control the automatic transparent replace ment of TCP sockets with SDP sockets In this mode socket streams are converted based upon a destination port a listening port or a program name Socket control statements in 1ibsdp conf allow the user to specify when 1ibsdp should replace AF INET SOCK STREAM sockets with AF SDP SOCK STREAM sockets Each control state ment specifies a matching rule that applies 1f all its subexpressions must evaluate as true logical and The use statement controls which type of sockets to open The format of a use statement is as fol lows use lt address family gt
24. RDS SRP The following is an example of QoS configuration for an enterprise data center EDC with IPoIB carrying all application traffic RDS for database traffic and SRP used for storage QoS Levels Management traffic ssh IPoIB management VLAN partition A Min BW 10 Application traffic IPoIB application VLAN partition B Isolated from storage and database Min BW of 30 Database Cluster traffic RDS Min BW of 30 SRP Min BW 30 Bottleneck at storage nodes Administration OpenSM QoS policy file Note In the following policy file example replace SRPT with the real SRP Initiator port GUIDs qos ulps default 0 ipoib pkey 0x8001 1 ipoib pkey 0x8002 22 rds 3 srp target port guid SRPT1 SRPT2 SRPT3 4 end qos ulps OpenSM options file qos max vis 8 qos high limit 0 qos vlarb high 1 32 2 96 3 96 4 96 qos vlarb low 0 1 qos sl2vl 0 1 2 3 4 5 0 7 15 15 15 15 15 15 15 15 Partition configuration file Default 0x7fff ipoib ALL full PartA 0x8001 sl 1 ipoib ALL full PartB 0x8002 sl 2 ipoib ALL full Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 11 1 11 2 11 2 1 InfiniBand Fabric Diagnostic Utilities Overview The diagnostic utilities described in this chapter provide means for debugging the connectivity and status of InfiniBand IB devices in a fabric The tools are ibdiagnet IB Net Diagnostic page
25. This command shows all active SDP sockets using the same format as the traditional netstat pro gram Without the S option it shows all the information that netstat does plus SDP data Assuming that the SDP kernel module is loaded and is being used then the output of the command will bve as follows hostl sdpnetstat S Proto Recv Q Send Q Local Address Foreign Address sdp sdp 0 0193 168 10 144 34216 193 168 10 125 12865 0 884720 193 168 10 144 42724 193 168 10 filenet rmi The example output above shows two active SDP sockets and contains details about the connec tions If the SDP kernel module is not loaded then the output of the command will be something like the following hostl sdpnetstat S Proto Recv Q Send Q Local Address Foreign Address netstat no support for AF INET tcp on this system To verify whether the module is loaded or not you can use the 1smod command hostl lsmod grep sdp ib sdp 125020 0 6 3 2 The example output above shows that the SDP module is loaded If the SDP module is loaded and the sapnetstat command did not show SDP sockets then SDP is not being used by any application Monitoring and Troubleshooting Tools SDP has debug support for both the user space 1ibsdp so library and the ib s p kernel module Both can be useful to understand why a TCP socket was not redirected over SDP and to help find problems in the SDP implementation User Space SDP Debug User sp
26. Tool option d This option defines a directed route of output port numbers from the local port to the destination Using port LIDs Tool option 1 In this mode the source and destination ports are defined by means of their LIDs If the fabric is config ured to allow multiple LIDs per port then using any of them is valid for defining a port Using port names defined in the topology file Tool option n This option refers to the source and destination ports by the names defined in the topology file There fore this option is relevant only if a topology file is specified to the tool In this mode the tool uses the names to extract the port LIDs from the matched topology then the tool operates as in the l option ibdiagnet IB Net Diagnostic ibdiagnet scans the fabric using directed route packets and extracts all the available information regarding its connectivity and devices It then produces the following files in the output directory which is defined by the o option described below 11 3 1 SYNOPSYS ibdiagnet c lt count gt v r o lt out dir gt t lt topo file gt s lt sys name gt i lt dev index gt p lt port num gt wt pm pc P lt lt PM gt lt Value gt gt lw lt 1 4 12 gt ls lt 2 5 5 10 gt Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 119 skip ibdiag check s load_db db file
27. sbin Ispci d 15b3 634a 04 00 0 InfiniBand Mellanox Technologies MT25418 ConnectX IB DDR PCIe 2 0 2 5GT s rev a0 In the example above 15b3 is Mellanox Technologies s vendor number in hexadecimal and 634a is the device s PCI Device ID in hexadecimal The number string 04 00 0 identifies the device in the form bus dev fn Note The PCI Device IDs of Mellanox Technologies devices can be obtained from the PCI ID Repository Website at http pci ids ucw cz read PC 15b3 2 Verify the ConnectX firmware using its ID using the results of the example above gt mstflint d 04 00 0 v ConnectX failsafe image Start address 80000 Chunk size 80000 NOTE The addresses below are contiguous logical addresses Physical addresses on flash may be different based on the image start address and chunk size 0x00000038 0x000010db 0x0010a4 BOOT2 OK 0x000010dc 0x00004947 0x00386c BOOT2 OK 0x00004948 0x000052c7 0x000980 Configuration OK 0x000052c8 0x0000530b 0x000044 GUID OK 0x0000530c 0x0000542f 0x000124 Image Info OK 0x00005430 0x0000634f 0x000f20 DDR OK 0x00006350 0x0000f29b 0x008f4c DDR OK 0x0000f29c 0x0004749b 0x038200 DDR OK 0x0004749c 0x0005913f 0x011ca4 DDR OK 0x00059140 0x0007a123 0x020fe4 DDR OK 0x0007a124 0x0007bdff 0x001cdc DDR OK 0x0007be00 0x0007eb97 0x002d98 DDR OK 0x0007eb98 0x0007f0af 0x000518 Configuration OK 0x0007f
28. t Test Name Specify the test to perform Options are TCP STREAM TCP RR etc C Client CPU utilization C Server CPU utilization Separates the global and test specific parameters m Message size which is 65536 in the example above Note that the run example above produced the following results Throughput is 2 483 gigabits per second e Client CPU utilization is 7 03 percent of client CPU Server CPU utilization is 5 42 percent of server CPU Step 5 Run the Netperf Latency test Run the test once and stop the server so that it does not repeat the test The following example shows how to run the Latency test and then stop the Netperf server host2 netperf H 11 4 17 6 t TCP RR c C r1 1 TCP REQUEST RESPONSE TEST from 0 0 0 0 0 0 0 0 port 0 AF INET to 11 4 17 6 11 4 17 6 port 0 AF INET Local Remote Socket Size Request Resp Elapsed Trans CPU CPU S dem S dem Send Recv Size Size Time Rate local remote local remote bytes bytes bytes bytes secs per sec S 5 us Tr us Tr 16384 87380 1 1 10 00 19913 18 5 61 6 79 22 549 21 296 16384 87380 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual The following table describes parameters for the netperf command Option Description H Where to find the server 11 4 17 6 IPoIB IP address t lt Test Name gt Specify the test to perform Options are TCP STREAM TCP RR etc C Clien
29. which have unique different VLAN IDs Each vHub belongs to a specific GW BridgeX eport and each GW has one default vHub and zero or more VLAN associated vHubs A specific GW can have multiple vHubs distinguishable by their unique virtual LAN VLAN id Traffic coming from the Ethernet side on a specific eport will be routed to the relevant vHub group based on its VLAN tag or to the default vHub for that GW if no VLAN ID is present Virtual NIC vNic A virtual NIC is a network interface instance on the host side The vNic behaves like regular HW network interface A vNic belongs to a single vHub on a specific GW The host can have multiple interfaces that belong to the same vHub Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 4 3 4 3 1 Figure 3 eports vNics and vHubs Ethernet port Ethernet port Ethernet port eport BridgeX VLAN id yHub Different EoIB Configuration The mlx4 vnic module supports two different modes of configuration host administration and net work administration In the first the vNic is configured on the host side and in the latter the config uration is done by the BridgeX and this configuration is passed to the host mlx4_vnic driver using the EoIB protocol Both modes of operation require the presence of a BridgeX gateway in order to work properly The EoIB driver supports a mixture
30. 0 open Link down TX 0 TXE O RX 0 RXE 0 Waiting for link up on 0 ok J HCP netO 00 02 c9 00 01 77 70 51 ok 1et0 11 4 3 130 255 255 255 0 gw 0 0 0 0 For InfiniHost III Ex BPXE 0 9 3 Open Source Boot Firmware features TFIP iSCSI AoE PXE PXEXI 1 10 00550401 fe800000 00000000 0002c902 00231392 on PCI05 00 0 0 RX 0 RXE 0 DHCP netO 00550401 800000 00000000 0002c902 002313927 ok 1etO 11 4 3 130 255 Next BoIB attempts to boot as directed by the DHCP server if open TX Mellanox Technologies Rev 1 10 156 A 8 Diskless Machines Mellanox Boot over IB supports booting diskless machines To enable using an IB driver the remote kernel or initrd image must include and be configured to load the IB driver including IPoIB This can be achieved either by compiling the HCA driver into the kernel or by adding the device driver module into the initrd image and loading it The IB driver requires loading the following modules in the specified order see Section A 8 1 for an example ib addr ko ib core ko ib mad ko ib sa ko ib_cm ko ib uverbs ko ib ucm ko ib umad ko iw cm ko rdma cm ko rdma ucm ko mlx4 core ko mlx4 ib ko ib mthca ko ib ipoib ko A 8 1 Example Adding an IB Driver to initrd Linux Prerequisites 1 The BoIB image is already programmed on the HCA card 2 The DHCP server is installed and configured as described
31. 0x decimal numbers are accepted too full or limited indicates full or limited membership for this port When omitted or unrecognized limited membership is assumed There are two useful keywords for PortGUID definition ALL means all end ports in this subnet SELF means subnet manager s port An empty list means that there are no ports in this partition Notes e White space is permitted between delimiters 5 The line can be wrapped after after a Partition Definition and between A PartitionName does not need to be unique but PKey does need to be unique f a PKey is repeated then the associated partition configurations will be merged and the first PartitionName will be used see also next note It is possible to split a partition configuration in more than one definition but then they PKey should be explicitly specified otherwise different PKey values will be generated for those defi nitions Examples Default 0x7fff ALL SELF full NewPartition ipoib 0x123456 full 0x3456789034 limi 0x2134af2306 YetAnotherOne 0x300 SELF full YetAnotherOne 0x300 ALL limited ShareIO 0x80 defmember full 0x123451 0x123452 0x123453 0x123454 will be limited ShareIO 0x80 0x123453 0x123454 0x123455 full 0x123456 0x123457 will be limited SharelO 0x80 defmember limited 0x123456 0x123457 0x123458 full ShareIO 0x80 defmember full 0x123459 0x
32. 118 ibdiagpath IB diagnostic path page 121 e ibv devices page 122 ibv devinfo page 123 ibstatus page 124 ibportstate page 126 ibroute page 131 smpquery page 135 perfquery page 138 ibcheckerrs page 141 mstflint page 143 e jbv asyncwatch page 147 Utilities Usage This section first describes common configuration interface and addressing for all the tools in the package Then it provides detailed descriptions of the tools themselves including operation synop sis and options descriptions error codes and examples Common Configuration Interface and Addressing Topology File Optional An InfiniBand fabric is composed of switches and channel adapter HCA TCA devices To iden tify devices in a fabric or even in one switch system each device is given a GUID a MAC equiv alent Since a GUID is a non user friendly string of characters it is better to alias it to a meaningful user given name For this objective the IB Diagnostic Tools can be provided with a topology file which is an optional configuration file specifying the IB fabric topology in user given names For diagnostic tools to fully support the topology file the user may need to provide the local system name if the local hostname is not used in the topology file To specify a topology file to a diagnostic tool use one of the following two options 1 On the command line specify the file name using the option t topo
33. 5GT s 25408 0x6340 fw 25408 MT25408 ConnectX IB DDR PCI Express 2 0 2 5GT s 25418 0x634a fw 25408 MT25408 ConnectX IB DDR PCI Express 2 0 5 0GT s 26418 0x6732 fw 25408 MT25408 ConnectX IB QDR PCI Express 2 0 5 0GT s 26428 0x673c fw 25408 MT25208 InfiniHost III Ex 25218 0x6282 fw 25218 MT25204 InfiniHost III Lx 25204 0x6274 fw 25204 Tested Platforms See Mellanox ConnectX EN PXE Release Notes ConnectX EN PXE release notes txt ConnectX EN PXE in Mellanox BXOFED The ConnectX EN PXE binary files are provided as part of the Mellaox BXOFED for Linux ISO image The following files are included 1 A PXE ROM image file for each of the supported Mellanox network adapter devices For exam ple ConnectX EN PCI DevID 25448 Mellanox Technologies Rev 1 10 176 B 2 CONNECTX EN 25448 ROM version rom Burning the Expansion ROM Image The binary code resides in the same Flash device of the device firmware Note that the binary files are distinct and do not affect each other Mellanox s m1xburn tool is available for burning how ever it is not possible to burn the expansion ROM image by itself Rather both the firmware and expansion ROM images must be burnt simultaneously mlxburn requires the following items 1 MST device name After installing the MFT package run mst start mst status B 3 B 3 1 The device name will be of the form dev mst mt dev
34. 7 A 10 WinPE Mellanox BoIB enables WinPE boot via TFTP For instructions on preparing a WinPE image please see http etherboot org wiki winpe Mellanox Technologies Rev 1 10 174 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 175 Appendix B ConnectX EN PXE B 1 B 1 1 B 1 2 B 1 3 Overview This appendix describes ConnectX EN PXE the software for Boot over Mellanox Technologies network adapter devices supporting Ethernet Mellanox ConnectX EN PXE enables booting kernels or operating systems OSes from remote servers in compliance with the PXE spec ification Mellanox ConnectX EN PXE is based on the open source project Etherboot gPXE available at http www etherboot org Mellanox ConnectX EN PXE first initializes the network adapter device Then it connects to a DHCP server to obtain its assigned IP address and network parameters and also to obtain the source location of the kernel OS to boot from The DHCP server instructs Mellanox ConnectX EN PXE to access the kernel OS through a TFTP server an iSCSI target or other service The binary code is exported by the device as an expansion ROM image Supported Mellanox Network Adapter Devices and Firmware Table 36 Supported Mellanox Technologies Devices and PCI Device IDs Device Name Firmware Name MT25408 ConnectX IB SDR PCI Express 2 0 2
35. C 2 InfiniBand Performance Troubleshooting 197 Appendix D ULP Performance Tuning 199 D 1 IPoIB Performance Tuning 199 D 2 Ethernet Performance Tuning 199 D 3 MPI Performance Tuning 200 Appendix E SRP Target Driver 201 E l Prerequisites 201 E 2 How to run 203 E 3 How to Unload Shutdown 206 Appendix F mlx4 Module Parameters 207 F 1 mlx4 core Parameters 207 2 mlx4 ib Parameters 208 F 3 mlx4 en Parameters 208 4 mlx4 fc Parameters 208 Glossary 209 Mellanox Technologies Rev 1 10 8 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 9 List of Tables Table 1 Typographical Conventions 12 Table 2 Abbreviations and Acronyms 13 Table 3 Reference Documents 14 Table 4 Supported ConnectX Port Configurations 21 Table 5 Useful MPI Links 73 Table 6 ibdiagnet Output Files 120 Table 7 ibdiagpath Output Files 122 Table 8 ibv devinfo Flags and Options 123 Table 9 ibstatus Flags and Options 125 Table 10 ibportstate Flags and Options 127 Table 11 ibportstate Flags and Options 131 Table 12 smpquery Flags and Options 135 Table 13 perfquery Flags and Options 139 Table 14 ibcheckerrs Flags and Options 142 Table 15 mstflint Switches 144 Table 16 mstflint Commands 145 Table 17 Supported Mellanox Technologies Devices and PCI Device IDs 149 Table 18 Supported Mellanox Technologies Devices and PCI Device IDs 175 Mellanox Technologies Rev 1 10 10 Rev 1 10 Mellanox Technolog
36. D 2 IPoIB Performance Tuning This section provides tuning guidelines of TCP stack configuration parameters in order to boost IPoIB and IPoIB CM performance Without tuning the parameters the default Linux configuration may significantly limit the total available bandwidth below the actual capabilities of the adapter card The parameter settings described below will increase the ability of Linux to transmit and receive data Generally if you increase the MTU maximum transmission unit in bytes you get better perfor mance The following MTUs are suggested use ifconfig to modify the MTU IPoIB 2044 bytes IPoIB CM 64K bytes When IPoIB is configured to run in connected mode TCP parameter tuning is performed at driver startup to improve the throughput of medium and large messages The driver startup scripts set the following TCP parameters as follows Note The following settings should not be applied when running in datagram mode as they degrade the performance net ipv4 tcp_timestamps 0 net ipv4 tcp_sack 0 net core netdev_max_backlog 250000 net core rmem_max 16777216 net core wmem max 16777216 net core rmem default 16777216 net core wmem default 16777216 net core optmem max 16777216 netipv4 tcp mem 16777216 16777216 16777216 net ipv4 tcp_rmem 4096 87380 16777216 net ipv4 tcp_ wmem 4096 65536 16777216 If you change the IPoIB run mode to datagram while the driver is running then the tuned parameters do not get
37. Destination Port gies 0x0003 021 Switch portguid 0x000b8cffff004016 MT47396 Infiniscale III Mellanox Technolo gies Info 0x0002 000 Switch portguid 0x0002c902fffff00a MT47396 Infiniscale III Mellanox Technolo 0x0006 007 Channel Adapter portguid 0x0002c90300001039 sw137 HCA 1 0x0007 021 Channel Adapter portguid 0x0002c9020025874a sw157 HCA 1 0x0008 008 Channel Adapter portguid 0x0002c902002582cd sw136 HCA 1 5 valid lids dumped Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 133 2 Dump all Lids with valid out ports of the switch with Lid 2 gt ibroute 2 Unicast lids 0x0 0x8 of switch Lid 2 guid 0x0002c902fffff00a MT47396 Infiniscale III Mel lanox Technologies Lid Out Destination Port Info 0x0002 000 Switch portguid 0x0002c902fffff00a MT47396 Infiniscale III Mellanox Technolo gies 0x0003 021 Switch portguid 0x000b8cffff004016 MT47396 Infiniscale III Mellanox Technolo gies 0x0006 007 Channel Adapter portguid 0x0002c90300001039 sw137 HCA 1 0x0007 021 Channel Adapter portguid 0x0002c9020025874a sw157 HCA 1 0x0008 008 Channel Adapter portguid 0x0002c902002582cd sw136 HCA 1 5 valid lids dumped 3 Dump all Lids in the range 3 to 7 with valid out ports of the switch with Lid 2 gt ibroute 2 3 7 Unicast lids 0x3 0x7 of switch Lid 2 guid 0x0002c902fffff00a MT47396 Infiniscale III Mel lanox
38. Device host ch id lun or name Device handler 0 0 0 0 dev disk 4 0 0 0 dev disk 5 0 0 0 dev disk 6 0 0 0 dev disk 7 0 0 0 dev disk Now you want to exclude the first scsi disk and expose the last 4 scsi disks as IB SRP luns for I O echo add 4 0 0 0 0 gt proc scsi_tgt groups Default devices echo add 5 0 0 0 1 gt proc scsi_tgt groups Default devices echo add 6 0 0 0 2 gt proc scsi_tgt groups Default devices echo add 7 0 0 0 3 gt proc scsi_tgt groups Default devices Example 3 working with scst vdisk FILEIO mode Using md0 device and file 10G file modprobe scst modprobe scst_vdisk echo open vdisk0 dev md0 gt proc scsi_tgt vdisk vdisk echo open vdisk1 10G file gt proc scsi_tgt vdisk vdisk echo add vdisk0 0 gt proc scsi_tgt groups Default devices echo add vdisk1 1 gt proc scsi_tgt groups Default devices 2 modprobe ib srpt B On Initiator Machines On Initiaor machines you can manualy do the following steps 1 modprobe ib srp 2 ipsrpdm c d dev infiniband umadX to discover new SRP target umado port 1 of the first HCA umad1 port 2 of the first HCA Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 205 umad2 port 1 of the second HCA 3 echo new target info gt sys class infiniband_srp srp mthca0 1 add_target 4 fdisk 1 will show the newly discovered scsi disks Example Assume that you use port 1 of first HCA in the system i e mthc
39. Std 802 3 2002 Document PDF 5594996 Part 3 Carrier Sense Multiple Access with Collision Detection CSMA CD Access Method and Physical Layer Specifications Amendment Media Access Control MAC Parameters Physical Layers and Management Parameters for 10 Gb s Operation Fibre Channel BackBone 5 standard for Fibre Channel over Ethernet Document INCITS xxx 200x Fibre Channel Backbone http www t11 org draft BridgeX Programmer s Reference Manual Document 2936PM Describes the software interface used by developers to write a driver for Mellanox BridgeX devices Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 1 1 1 1 2 1 3 1 3 1 Mellanox BXOFED Overview Introduction to Mellanox BXOFED BXOFED or OFED with BridgeX support is a single Virtual Protocol Internconnect VPI soft ware stack based on the OpenFabrics OFED Linux stack and operates across all Mellanox net work adapter solutions supporting 10 20 and 40Gb s InfiniBand IB 10Gb s Ethernet 10GigE Fibre Channel over Ethernet FCoE Fibre Channel over InfiniBand FCoIB connected via Mel lanox BridgeX gateways Ethernet over InfiniBand EoIB connected via Mellanox BridgeX gateways and 2 5 or 5 0 GT s PCI Express 2 0 uplinks to servers All Mellanox network adapter cards are compatible with OpenFabrics based RDMA protocols and software and are supported with major operating sy
40. Technologies Lid Out Destination Port Info 0x0003 021 Switch portguid 0x000b8cffff004016 MT47396 Infiniscale III Mellanox Technolo gies 0x0006 007 Channel Adapter portguid 0x0002c90300001039 sw137 HCA 1 0x0007 021 Channel Adapter portguid 0x0002c9020025874a sw157 HCA 1 3 valid lids dumped Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 134 4 Dump all Lids with valid out ports of the switch with portguid 0x000b8c 004016 gt ibroute G 0x000b8cffff004016 Unicast lids 0x0 0x8 of switch Lid 3 guid 0x000b8cffff004016 MT47396 Infiniscale III Mel lanox Technologies Lid Out Destination Port Info 0x0002 023 Switch portguid 0x0002c902fffff00a MT47396 Infiniscale III Mellanox Technolo gies 0x0003 000 Switch portguid 0x000b8cffff004016 MT47396 Infiniscale III Mellanox Technolo gies 0x0006 023 Channel Adapter portguid 0x0002c90300001039 sw137 HCA 1 0x0007 020 Channel Adapter portguid 0x0002c9020025874a sw157 HCA 1 0x0008 024 Channel Adapter portguid 0x0002c902002582cd sw136 HCA 1 5 valid lids dumped 5 Dump all non empty mlids of switch with Lid 3 gt ibroute M 3 Multicast mlids 0xc000 0xc3 ff of switch Lid 3 guid 0x000b8cffff004016 MT47396 Infiniscale III Mellanox Technologies 0 1 p Ports 0123456789012345678901234 MLid 0xc000 0xc001 0xc002 0xc003 0xc020 0xc021 0xc022 0xc023 0xc024 0xc040 0xc041 0xc042
41. True for this target Click Next See figure below Mellanox Technologies Rev 1 10 164 Preparation Vv Language License Agreement gt Disk Activation e System Analysis Connected e Time Zone 10 4 3 3 oot True iSCSI Initiator Discovery Installation e Installation Summary e Perform Installation Configuration e Root Password Hostname e Network e Customer Center e Online Update e Service e Users e Clean Up e Release Notes e Hardware Configuration Step 7 The iSCSI Initiator Overview window will pop up Click Toggle Start Up to change start up from manual to automatic Click Finish Preparation Language License Agreement gt Disk Activation e System Analysis Service Connected Targets e Time Zone Portal Address s Target Name Start Up di Installation 10 4 3 7 60 1 iqn 2007 3 4 10 manual e Installation Summary e Perform Installation iSCSI Initiator Overview Configuration e Root Password e Hostname e Network e Customer Center e Online Update e Service e Users e Clean Up e Release Notes e Hardware Configuration Add Log Out Toggle Start Up Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual Step 8 Select New Installation then click Finish in the Installation Mode window Preparation v Language License Agreement Disk Activat
42. a higher limit of packet rate adaptive interrupt moderation will set the moderation time to its high est value below a lower limit of packet rate adaptive interrupt moderation will set the moderation time to its lowest value To set the values for packet rate limits and moderation time high low values use the following command gt ethtool C eth lt n gt pkt rate low N pkt rate high N rx usecs low N rx usecs high N To set interrupt coalescing settings when adaptive moderation is disabled run gt ethtool c eth lt n gt rx usecs N rx frames N Note usec settings correspond to the time to wait after the last packet sent received before triggering an interrupt To query pause frame settings run Mellanox Technologies Rev 1 10 23 Working With VPI 24 ethtool a eth lt n gt To set pause frame settings run gt ethtool A eth lt n gt rx onloff tx on off To obtain additional device statistics run gt ethtool S eth lt n gt The mlx4 en parameters can be found under sys module mlx4 en or sys module mlx4 en parameters depending on the OS and can be listed using the command gt modinfo mlx4 en To set non default values to module parameters the following line should be added to the file etc modprobe conf options mlx4 en param name gt lt value gt param name gt lt value gt 24 Fibre Channel over Ethernet 2 4 1 Overview The FCoE feature p
43. a partition on your iSCSI Target on which you will later install the operating sys tem Mellanox Technologies Rev 1 10 160 Step 3 Configure your iSCSI Target to work with the partition you dedicated If for example you choose partition dev sda5 then edit the iSCSI Target configuration file etc ietd conf to include the following line under the iSCSI Target iqn line Lun 0 Path dev sda5 Type fileio Tip The following is an example of an iSCSI Target iqn line Target iqn 2007 08 7 3 4 10 iscsiboot Step 4 Start your iSCSI Target Example host1 etc init d iscsitarget start Configuring the DHCP Server to Boot From an iSCSI Target Configure DHCP as described in Section A 3 1 Configuring the DHCP Server Edit your DHCP configuration file etc dhcpd conf and add the following lines for the machine s you wish to boot from the iSCSI Target nn A Filename option root path 15 51 15 51 target 1 15 51 target iqn The following is an example for configuring an IB device to boot from an iSCSI Target host hostl filename For a ConnectX device comment out the following line option dhcp client identifier 00 02 c9 03 00 00 10 39 For an InfiniHost III Ex comment out the following line option dhcp client identifier fe 00 55 00 41 fe 80 00 00 00 00 00 00 00 02 c9 03 00 00 0d 41 option root path 1scsi 11 4 3 7 10n 2007 08 7 3 4 10 1scsiboot j A 9 2 iSCSI Boot Exampl
44. any query that has this Ser vice ID disregarding rest of the query fields However if a certain query has only Service ID which means that this is the only bit in the PR MPR component mask that is on it will not match any rule that has other matching criteria besides Service ID 10 6 3 Simple QoS Policy Definition Simple QoS policy definition comprises of a single section denoted by qos ulps Similar to the advanced QoS policy it has a list of match rules and their QoS Level but in this case a match rule Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 10 6 4 10 6 5 has only one criterion its goal is to match a certain ULP or a certain application on top of this ULP PR MPR request and QoS Level has only one constraint Service Level SL The simple policy section may appear in the policy file in combine with the advanced policy or as a stand alone policy definition See more details and list of match rule criteria below Policy File Syntax Guidelines Leading and trailing blanks as well as empty lines are ignored so the indentation in the exam ple is just for better readability Comments are started with the pound sign and terminated by EOL Any keyword should be the first non blank in the line unless it s a comment Keywords that denote section subsection start have matching closing keywords Having a QoS Level named DEFAULT is a must it is applied to PR M
45. as follows node description Pnum node description is compared to the NodeDescription of the node and Pnum is a port number on that node port name vs1 HCA 1 P1 vs2 HCA 1 P1 end port group using partitions defined in the partition policy port group name Partitions partition Part1 pkey 0x1234 end port group using node types CA ROUTER SWITCH SELF for node that runs SM or ALL for all the nodes in the subnet port group name CAs and SM node type CA SELF end port group end port groups qos setup This section of the policy file describes how to set up SL2VL and VL Arbitration tables on various nodes in the fabric However this is not supported in BXOFED the section is parsed and ignored SL2VL and VLArb tables should be configured in the OpenSM options file by default var cache opensm opensm opts end qos setup qos levels Having a QoS Level named DEFAULT is a must it is applied to PR MPR requests that didn t match any of the matching rules qos level name DEFAULT use default QoS Level sl 0 end qos level the whole set SL MTU Limit Rate Limit PKey Packet Lifetime qos level name WholeSet sl 1 mtu limit 4 rate limit 5 pkey 0x1234 packet life 8 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 109 end qos level end qos levels Match rules are scanned in order of their apperance in the polic
46. bondO IP 10 10 10 1 24 bond0_SLAVES ib0 ib1 bond8007_IP 20 10 10 1 bond1_SLAVES ib0 8007 ib1 8007 b Restart the driver by running Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 3 7 3 8 35 etc init d openibd restart 2 Using a standard OS bonding configuration For details on this please read the documentation for the ib bonding package under usr share doc ib bonding 0 9 0 ib bonding txt on RedHat and usr share doc packages ib bonding 0 9 0 ib bonding txt on SuSE Notes If the bondX name is defined but one of bondX SLAVES or bondX IPs is missing then that specific bond will not be created The bondX name must not contain characters which are disallowed for bash variable names such as and Using etc infiniband openib conf fo create a persistent configuration is not recom mended Do not use it unless you have no other option It is not guaranteed that the first method will be supported in future versions of BXOFED IPoIB Performance Tuning When IPoIB is configured to run in connected mode TCP parameter tuning is performed at driver startup to improve the throughput of medium and large messages Testing IPoIB Performance This section describes how to verify IPoIB performance by running the Bandwidth BW test and the Latency test These tests are described in detail at the following URL http www netperf org netperf training Netperf html Not
47. client v3 1 2 needs to be already installed on the machine you are working with Copy the DHCP client v3 1 2 file and all the relevant files as described below hostl cp path to DHCP client v3 1 2 gt dhclient tmp initrd ib sbin Mellanox Technologies Rev 1 10 158 hostl cp path to DHCP client v3 1 2 gt dhclient script tmp initrd ib sbin hostl mkdir p tmp initrd ib var state dhcp hostili touch tmp initrd ib var state dhcp dhclient leases hostl cp bin uname tmp initrd ib bin hostl cp usr bin expr tmp initrd ib bin hostl cp sbin ifconfig tmp initrd ib bin hostl cp bin hostname tmp initrd ib bin Create a configuration file for the DHCP client as described in Section 3 3 1 2 and place it under tmp initrd ib sbin The following is an example of such a file called dclient conf dhclient conf The value indicates a hexadecimal number For a ConnectX device interface 160 send dhcp client identifier 00 02 c9 03 00 00 10 39 For an InfiniHost III Ex device interface ib1 send dhcp client identifier 20 00 55 04 01 fe 80 00 00 00 00 00 00 00 02 c9 02 00 23 13 92 Step 9 Now you can add commands for loading the copied modules into the file init Edit the file tmp initrd ib init and add the following lines at the point you wish the IB driver to be loaded Warning The order of the following commands for loading modules is critical echo loading ipv6 sbin i
48. for advanced options like RAID and LVM Partitioning Accept Proposal Base Partition Setup on This Proposal Create Custom Partition Setup Back Abort Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 167 Step 12 In the Expert Partitioner window select from the IET VIRTUAL DISK device the row that has its Mount column indicating swap then click Delete Confirm the delete opera tion and click Finish Partition your hard Expert Partitioner disks This is intended for experts If you are not Device Size Type Mount Mount By Start End Used By Label Devi Tamiliarw hie Jdev sda 8 0GB IETVIRTUALDISK 0 1045 scsi conceptsuthard disk Jdev sdal 70 5 8 Linux native Ext2 boot 0 8 scsi partitiongend haw c dev sda2 502 0 MB F Linux swap swap use them you might z wantto go back and Idev sda3 F Linux native Reiser 73 1045 select automatic partitioning Please note that nothing will be written to your hard disk until you confirm Really delete device dev sda2 the entire installation in the last installation dialog Until that point you can safely abort the installation For LVM setup using a non LVM root device and a non LVM swap device is recommended Other than the root and swap devices you should have partitions managed by LVM
49. full advantage of the protocol offload and RDMA features provided by the InfiniBand architecture SRP allows a large body of SCSI software to be readily used on InfiniBand architecture The SRP Initiator con trols the connection to an SRP Target in order to provide access to remote storage devices across an InfiniBand fabric The SRP Target resides in an IO unit and provides storage services Section 7 2 describes the SRP Initiator included in Mellanox BXOFED for Linux This package however does not include an SRP Target SRP Initiator This SRP Initiator is based on open source from OpenFabrics www openfabrics org that imple ments the SCSI Protocol 2 SRP 2 SRP 2 is described in Document T10 1524 D avail able from www tl0 org ftp t10 drafts srp2 srp2r00a pdf The SRP Initiator supports Basic SCSI Primary Commands 3 SPC 3 www t10 org ftp tl O drafts spc3 spc3r21b pdf Basic SCSI Block Commands 2 SBC 2 www t10 org ftp tl0 drafts sbc2 sbc2r16 pdf Basic functionality task management and limited error handling Loading SRP Initiator To load the SRP module either execute the modprobe ib srp command after the BXOFED driver is up or change the value of SRP LOAD in etc infiniband openib conf to yes Note For the changes to take effect run etc init d openibd restart Note When loading the ib srp module it is possible to set the module parameter srp sg tablesize This is the maximum n
50. gt OPTIONS c count t topo file s lt sys name gt i dev index p lt port num gt o out dir lw 1x 4x 12x ATS 2 5J5 2T102 pm pc P lt PM lt Trash gt gt in number of packets to be sent across each link default 10 Enable verbose mod Provides a report of the fabric qualities pecifies the topology file name pecifies the local system name Meaningful only if topology file is specified pecifies the index of the device of the port used to onnect to the IB fabric in case of multiple evices on the local system ect to the IB fabric pecifies the directory where the output files will placed default tmp pecifies th xpected link width S S a S d Specifies the local device s port num used to con n S b S S pecifies th xpected link speed Dump all the fabric links pm Counters into ibdiag Reset all the fabric links pmCounters If any of the provided pm is greater then its pro vided value print it to screen skip lt skip option s gt Skip the executions of the selected checks Skip wt file name options one or more can be specified dup guids zero guids pm logical state part ipoib all Write out the discovered topology into the given file This flag is useful if you later want to check for changes from the current state of the fabric A directory named ibdiag ibnl is also created by this opti
51. ifconfig eth3 55 up To set the map of skb priority 0 to the requested vlan priority e g 6 run gt vconfig set egress map eth3 55 0 6 To create the vHBA enter gt echo eth3 55 gt FCSYSFS create 2 4 4 3 Creating vHBAs That Use Link Pause The mix4 en Ethernet driver supports link pause by default To change this setting you can use the following command gt ethtool A eth x rx onloff tx onloff To create a VHBA run gt echo eth3 55 gt FCSYSFS create 27 Mellanox Technologies Rev 1 10 Working With VPI 28 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 3 3 1 3 2 3 3 29 IPoIB Introduction The IP over IB IPoIB driver is a network interface implementation over InfiniBand IPoIB encap sulates IP datagrams over an InfiniBand Connected or Datagram transport service This chapter describes the following PoIB mode setting Section 3 2 PoIB configuration Section 3 3 How to create and remove subinterfaces Section 3 4 How to verify IPoIB functionality Section 3 5 The ib bonding driver Section 3 6 PoIB performance tuning Section 3 7 How to test IPoIB performance Section 3 8 IPoIB Mode Setting IPoIB can run in two modes of operation Connected mode and Datagram mode By default IPoIB is set to work in Connected mode This can be changed to become Datagram mode by editing the file et
52. on the InfiniBand fabric at all times An SM can run on any node or on an IB switch OpenSM is an InfiniBand compliant Subnet Manager and it is installed as part of BXOFED See Chapter 10 OpenSM Subnet Manager Diagnostic Utilities Mellanox BXOFED includes the following two diagnostic packages for use by network and data center managers ibutils Mellanox Technologies diagnostic utilities infiniband diags OpenFabrics Alliance InfiniBand diagnostic tools Performance Utilities A collection of tests written over uverbs intended for use as a performance micro benchmark As an example the tests can be used for hardware or software tuning and or functional testing See PERF TEST README txt under docs Quality of Service Quality of Service QoS requirements stem from the realization of I O consolidation over an IB network As multiple applications and ULPs share the same fabric a means is needed to control their use of network resources QoS over Mellanox BXOFED for Linux is discussed in Chapter 10 OpenSM Subnet Manager 1 OpenSM is disabled by default See Chapter 10 OpenSM Subnet Manager for details on enabling it 19 Mellanox Technologies Rev 1 10 Mellanox BXOFED Overview 20 Rev 1 10 Mellanox Technologies 2 1 Working With VPI VPI allows ConnectX ConnectX 2 ports to be independently configured as either IB or Eth Ifa ConnectX port
53. or simply reset them Synopsys perfquery h d G a 1 r C ca name P ca port R t timeout ms V lt lid guid gt port reset mask Table 31 lists the various flags of the command Table 31 perfquery Flags and Options Fla Optional Default Description 8 Mandatory If Not Specified P h help Optional Print the help menu d ebug Optional Raise the IB debug level May be used several times for higher debug levels ddd or d d d G uid Optional Use GUID address argument In most cases it is the Port GUID Example 0x08f1040023 a Optional Apply query to all ports d Optional Loop ports r Optional Reset the counters after reading them C lt ca_name gt Optional Use the specified channel adapter or router P lt ca port gt Optional Use the specified port R Optional Reset the counters t lt timeout ms gt Optional Override the default timeout for the solicited MADs msec V ersion Optional Show version info lid guid gt Optional LID or GUID port reset mask Examples perfquery r 32 1 read performance counters and reset perfquery e r 32 1 read extended performance counters and reset perfquery R 0x20 1 Z reset performance counters of port 1 only perfquery e R 0x20 1 reset extended performance counters of port 1 only perfquery R 32 reset performance counters of all ports perfquery R 3
54. port GUID in the PR query so in order for these queries not to be recognized by the QoS manager as SRP the SRP match rule or any match rule that refers to the target port guid only should be placed at the end of the qos ulps match rules 10 6 6 6MPI SL for MPI is manually configured by MPI admin OpenSM is not forcing any SL on the MPI traf fic and that s why it is the only ULP that did not appear in the qos ulps section 10 6 7 SL2VL Mapping and VL Arbitration OpenSM cached options file has a set of QoS related configuration parameters that are used to con figure SL2VL mapping and VL arbitration on IB ports These parameters are Max VLs the maximum number of VLs that will be on the subnet High limit the limit of High Priority component of VL Arbitration table IBA 7 6 9 e VLArb low table Low priority VL Arbitration table IBA 7 6 9 template VLArb high table High priority VL Arbitration table IBA 7 6 9 template SL2VL SL2VL Mapping table IBA 7 6 6 template It is a list of VLs corresponding to SLs 0 15 Note that VL15 used here means drop this SL There are separate QoS configuration parameters sets for various target types CAs routers switch external ports and switch s enhanced port 0 The names of such parameters are prefixed by qos type string Here is a full list of the currently supported sets qos ca QoS configuration parameters set for CAs qos rtr parameters set for routers qos
55. so that the PKey chosen which defines a broadcast address be recognized see Chapter 10 OpenSM Subnet Manager Removing a Subinterface To remove a child interface subinterface run echo lt subinterface PKey gt sys class net lt ib interface gt delete child Using the example of Step 2 echo 0x8000 gt sys class net ib0 delete_child Note that when deleting the interface you must use the PKey value with the most significant bit set e g 0x8000 in the example above Verifying IPoIB Functionality To verify your configuration and your IPoIB functionality perform the following steps Step 1 Verify the IPoIB functionality by using the i config command The following example shows how two IB nodes are used to verify IPoIB functionality In the following example IB node 1 is at 11 4 3 175 and IB node 2 is at 11 4 3 176 host1 ifconfig ib0 11 4 3 175 netmask 255 255 0 0 host2 ifconfig 160 11 4 3 176 netmask 255 255 0 0 Step 2 Enter the ping command from 11 4 3 175 to 11 4 3 176 Mellanox Technologies Rev 1 10 34 IPoIB 3 6 3 6 1 a The following example shows how to enter the ping command hostl ping c 5 11 4 3 176 PING 11 4 3 176 11 4 3 176 56 84 bytes of data 64 bytes from 11 4 3 176 icmp_seq 0 ttl 64 time 0 079 ms 64 bytes from 11 4 3 176 icmp_seq 1 ttl 64 time 0 044 ms 64 bytes from 11 4 3 176 icmp seq 2 ttl 64 time 0 055 ms 64 bytes from 11 4 3 176 icmp_seq 3 ttl 64 ti
56. status This informa tion can be retrieved through the mlx4 vnic info script GW EPORT field value If the eport state enforce module parameter is set then the external port state will be reported as the vNic interface link state Naturally if the connection between the vNic and the BridgeX is broken and there for the external port state is unknown the link will be reported as down Note A link state of down on a host administrated vNics when the BridgeX is connected and the InfiniBand fabric seems OK is a good indication of a BridgeX system name or system GUID or eport mis configuration Check the value of BXADDR and BXEPORT in the configu ration file To query the link state run ifconfig interface name gt and check for RUNNING in the result text Example ifconfig eth2 eth2 Link encap Ethernet HWaddr 00 25 8B 00 04 00 inet6 addr fe80 225 8bff fe00 400 64 Scope Link UP BROADCAST RUNNING MULTICAST MTU 1500 Metric 1 RX packets 49 errors 0 dropped 11 overruns 0 frame 0 TX packets 25 errors 0 dropped 0 overruns 0 carrier 0 collisions 0 txqueuelen 1000 RX bytes 11278 11 0 KiB TX bytes 5821 5 6 KiB An alternative is to use ethtool lt interface name gt and test for Link detected Example Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 4 6 4 7 4 8 ethtool eth2 Settings for eth2 Supported ports Supported link modes Supports auto negotiation No Advertis
57. sure only one instance of run srp daemon runs per port To execute SRP daemon as a daemon on all the ports run srp daemon sh found under usr sbin srp daemon sh sends its log to var 1og srp daemon log Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 7 2 5 t is possible to configure this script to execute automatically when the InfiniBand driver starts by changing the value of SRPHA ENABLE in etc infiniband openib conf to yes However this option also enables SRP High Availability that has some more features see Sec tion 7 2 6 Note For the changes in openib conf to take effect run etc init d openibd restart Multiple Connections from Initiator IB Port to the Target Some system configurations may need multiple SRP connections from the SRP Initiator to the same SRP Target to the same Target IB port or to different IB ports on the same Target HCA In case of a single Target IB port i e SRP connections use the same path the configuration is enabled using a different initiator ext value for each SRP connection The initiator ext value is a 16 hexadecimal digit value specified in the connection command Also in case of two physical connections 1 network paths from a single initiator IB port to two different IB ports on the same Target HCA there is need for a different initiator ext value on each path The conventions is to use the Target port GUID as the ini
58. the LID2 Port 1 using the specified threshold file gt cat thresh SymbolErrors 10 LinkRecovers 10 LinkDowned 10 RevErrors 10 RcvRemotePhysErrors 100 RevSwRelayErrors 100 XmtDiscards 100 XmtConstraintErrors 100 RevConstraintErrors 100 LinkIntegrityErrors 10 ExcBufOverrunErrors 10 VL15Dropped 100 gt ibcheckerrs v T thresh 2 1 Error check on lid 2 MT47396 Infiniscale III Mellanox Technologies port 1 OK 11 13 mstflint Applicable Hardware Mellanox InfiniBand and Ethernet devices and network adapter cards Description Queries and burns a binary firmware image file on non volatile Flash memories of Mellanox InfiniBand and Ethernet network adapters The tool requires root privileges for Flash access Note If you purchased a standard Mellanox Technologies network adapter card please down load the firmware image from www mellanox com gt Downloads gt Firmware If you purchased a non standard card from a vendor other than Mellanox Technologies please contact your vendor To run mstflint you must know the device location on the PCI bus See Example 1 for details Synopsis mstflint switches lt command gt parameters Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 144 Table 33 lists the various switches of the utility and Table 34 lists its commands Table 33 mstflint Switches Sheet 1 of 2 Switc
59. to 11 4 17 6 11 4 17 6 port 0 AF INET Recv Send Send Utilization Service Demand Socket Socket Message Elapsed Send Recv Send Recv Size Size Size Time Throughput local remote local remote bytes bytes bytes secs 10 6bits s 8 S us KB us KB 87380 16384 65536 10 00 5872 60 19 41 17 12 2 166 1 911 Note You must specify the SDP IPoIB IP address when running the Netperf client Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual The following table describes parameters for the netperf command Option Description H Where to find the server 11 4 17 6 SDP IPoIB IP address t lt Test Name gt Specify the test to perform Options are TCP STREAM TCP RR etc C Client CPU utilization C Server CPU utilization Separates the global and test specific parameters m Message size which is 65536 in the example above Note that the run example above produced the following results Throughput is 5 872 gigabits per second Client CPU utilization is 19 41 percent of client CPU Server CPU utilization is 17 12 percent of server CPU Step 6 Run the Netperf Latency test such that you force SDP to be used instead of TCP Run the test once and stop the server so that it does not repeat the test host2 LD PRELOAD libsdp so LIBSDP CONFIG FILE HOME libsdp conf netperf V H 11 4 17 6 t RR c C r1 1 TCP REQUEST RESPONSE TEST from 0 0 0 0 0 0 0 0 port 0
60. to be typed by users as is bold font Keywords bold font Variables for which users supply specific values Italic font Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 13 Table 19 Typographical Conventions Description Convention Example Emphasized words Italic font These are emphasized words Pop up menu sequences menul gt menu2 gt gt item Note Note Warning Warning Common Abbreviations and Acronyms Table 20 Abbreviations and Acronyms Abbreviation Acronym Whole Word Description B Capital B is used to indicate size in bytes or multiples of bytes e g IKB 1024 bytes and IMB 1048576 bytes b Small b is used to indicate size in bits or multiples of bits e g IKb 1024 bits FCoE Fibre Channel over Ethernet FW Firmware HCA Host Channel Adapter HW Hardware IB InfiniBand LSB Least significant byte Isb Least significant bit MSB Most significant byte msb Most significant bit NIC Network Interface Card SW Software VPI Virtual Protocol Interconnect Mellanox Technologies Rev 1 10 14 Related Documentation Table 21 Reference Documents Document Name Description InfiniBand Architecture Specification Vol 1 Release 1 2 1 The InfiniBand Architecture Specification that is provided by IBTA IEEE Std 802 3ae 2002 Amendment to IEEE
61. used several times for additional verbosity vvv or v v v G uid Optional Use GUID address argument In most cases it is the Port GUID Example 0x08f1040023 T threshold file Optional Use specified threshold file S Optional Show the predefined thresholds N nocolor Optional color mode Use mono mode rather than color mode C lt ca_name gt Optional Use the specified channel adapter or router P lt ca port gt Optional Use the specified port t lt timeout ms gt Optional Override the default timeout for the solicited MADs msec lt lid guid Mandatory with G Use the specified port s or node s LID GUID with G option flag lt port gt Mandatory without Use the specified port G flag Examples 1 Check aggregated node counter for LID 0x2 gt ibcheckerrs 2 warn counter SymbolErrors 65535 threshold 10 lid 2 port 255 warn counter LinkRecovers 255 threshold 10 lid 2 port 255 warn counter LinkDowned 12 threshold 10 lid 2 port 255 warn counter RcvErrors 565 threshold 10 lid 2 port 255 warn counter XmtDiscards 441 threshold 100 lid 2 port 255 Error check on lid 2 MT47396 Infiniscale III Mellanox Technologies port all FAILED Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 143 2 Check port counters for LID 2 Port 1 gt ibcheckerrs v 2 1 Error check on lid 2 MT47396 Infiniscale III Mellanox Technologies port 1 OK 3 Check
62. 0 Path dev sda5 Type fileio Tip The following is an example of an iSCSI Target iqn line Target iqn 2007 08 7 3 4 10 iscsiboot Step 4 Start your iSCSI Target Example host1 etc init d iscsitarget start Configuring the DHCP Server to Boot From an iSCSI Target in Linux Environment Configure DHCP as described in Section B 3 1 Configuring the DHCP Server Edit your DHCP configuration file etc dhcpd conf and add the following lines for the machine s you wish to boot from the iSCSI Target Filename option root path 15 51 15 51 target 1 15 51 target iqn n The following is an example for configuring an Ethernet device to boot from an iSCSI Target host hostl filename hardware ethernet 00 02 c9 00 00 bb option root path iscsi 11 4 3 7 ign 2007 08 7 3 4 10 iscsiboot Mellanox Technologies Rev 1 10 182 2 9 iSCSI Boot Example of SLES 10 SP2 OS This section provides an example of installing the SLES 10 SP2 operating system on an iSCSI tar get and booting from a diskless machine via ConnectX EN PXE Note that the procedure described below assumes the following The client s LAN card is recognized during installation The iSCSI target can be connected to the client via a LAN and a ConnectX Ethernet Prerequisites See Section B 6 1 on page 178 Warning The following procedure modifies critical files used in the boot procedure It must be executed by users with ex
63. 000 2 67 256 1000 3 03 512 1000 3 64 1024 1000 4 89 2048 1000 6 30 4096 1000 8 91 8192 1000 14 07 16384 1000 18 85 32768 1000 30 47 65536 640 53 67 131072 320 99 78 262144 160 191 80 524288 80 373 92 1048576 40 74231 2097152 20 1475 20 4194304 10 2956 95 OUTPUT TRUNCATED 1352 75 8 6 Open MPI Performance 8 6 1 Requirements At least two nodes Example host1 host2 Machine file Includes the list of machines Example host1 cat home lt username gt cluster host1 host2 host1 8 6 2 Bandwidth Test Performance To run the OSU Bandwidth test enter Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual hostl usr mpi gcc openmpi lt ompi ver gt bin mpirun np 2 mca mpi leave pinned 1 hostfile home lt username gt cluster usr mpi gcc openmpi ompi ver tests osu benchmarks osu ver osu bw OSU MPI Bandwidth Test v3 0 H Size Bandwidth MB s 1 1 12 2 2 24 4 4 43 8 8 96 16 17 38 32 34 69 64 69 31 128 121 29 256 212 70 512 326 50 1024 461 78 2048 597 85 4096 543 06 8192 829 64 16384 1137 22 32768 1386 08 65536 1520 89 131072 1622 73 262144 1659 33 524288 1679 36 1048576 1675 35 2097152 1668 89 4194304 1671 78 8 6 3 Latency Test Performance To run the OSU Latency test enter hostl usr mpi gcc openmpi lt ompi ver gt bin mpirun np 2 mca mpi leave pinned 1 hostfile home lt username gt cluster usr mpi gcc openmpi
64. 0b0 0x0007f0fb 0x00004c Jump addresses OK 0x0007f0fc 0x0007f2a7 0x0001ac FW Configuration OK FW image verification succeeded Image is bootable Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 147 11 14 ibv_asynewatch Applicable Hardware All InfiniBand devices Description Display asynchronous events forwarded to userspace for an InfiniBand device Synopsis ibv_asyncwatch Examples 1 Display asynchronous events gt ibv_asyncwatch mlx4_0 async event FD 4 Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 148 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 149 Appendix A Boot over IB BoIB A 1 A 1 1 A 1 2 A 1 3 Overview This chapter describes Mellanox Boot over IB BoIB the software for Boot over Mellanox Technologies InfiniBand IB HCA devices BoIB enables booting kernels or operating systems OSs from remote servers in compliance with the PXE specification BolIB is based on the open source project Etherboot gPXE available at http www etherboot org BoIB first initializes the HCA device Then it connects to a DHCP server to obtain its assigned IP address and network parameters and also to obtain the source location of the kernel OS to boot from The DHCP server instructs BoIB to access the kernel OS through a TFTP server an iSCSI target or other serv
65. 12 valid mlids dumped WoW 0 P X Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 135 11 10 smpquery Applicable Hardware All InfiniBand devices Description Provides a basic subset of standard SMP queries to query Subnet management attributes such as node info node description switch info and port info Synopsys smpquery h d e v D G s lt smlid gt V C lt ca_name gt P lt ca_port gt t lt timeout_ms gt node name map lt node name map gt op dest dr path lid guid op params Table 30 lists the various flags of the command Table 30 smpquery Flags and Options Fla Optional Default Description 8 Mandatory If Not Specified h help Optional Print the help menu d ebug Optional Raise the IB debug level May be used several times for higher debug levels ddd or d d d e rr show Optional Show send and receive errors timeouts and others v erbose Optional Increase verbosity level May be used several times for additional verbosity vvv or v v v D irect Optional Use directed path address arguments The path is a comma sepa rated list of out ports Examples 0 self port 0 1 2 1 4 out via port 1 then 2 G uid Optional Use GUID address argument In most cases it is the Port GUID Example 0x08f1040023 s lt
66. 12345a SharelO 0x80 defmember full 0x12345b 0x12345c limited 0x12345d Note The following rule is equivalent to how OpenSM used to run prior to the partition man ager Default 0x 7fff ipoib ALL full 10 5 Routing Algorithms OpenSM offers five routing engines 1 Min Hop algorithm Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual vog 99 Based on the minimum hops to each node where the path length is optimized 2 UPDN Unicast routing algorithm Based on the minimum hops to each node but it is constrained to ranking rules This algorithm should be chosen if the subnet is not a pure Fat Tree and a deadlock may occur due to a loop in the subnet Fat Tree Unicast routing algorithm This algorithm optimizes routing for congestion free shift communication pattern It should be chosen if a subnet is a symmetrical Fat Tree of various types not just a K ary N Tree non constant not fully staffed and for any CBB ratio Similar to UPDN Fat Tree routing 1s constrained to ranking rules 4 LASH Unicast routing algorithm Uses InfiniBand virtual layers SL to provide deadlock free shortest path routing while also distributing the paths between layers LASH is an alternative deadlock free topology agnostic routing algorithm to the non minimal UPDN algorithm It avoids the use of a potentially congested root node DOR Unicast routing algorithm Based
67. 2 lt username gt host2 s password Enter password host1 For a local machine simply add the key to authorized keys2 host1 cat id rsa pub gt gt authorized keys2 Step 5 Test hostl ssh host2 uname Linux 8 3 MPI Selector Which MPI Runs Mellanox BXOFED contains a simple mechanism for system administrators and end users to select which MPI implementation they want to use The MPI selector functionality is not specific to any MPI implementation it can be used with any implementation that provides shell startup files that Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 8 4 75 correctly set the environment for that MPI The Mellanox BXOFED installer will automatically add MPI selector support for each MPI that it installs Additional MPI s not known by the Mellanox BXOFED installer can be listed in the MPI selector see the mpi selector 1 man page for details Note that MPI selector only affects the default MPI environment for future shells Specifically if you use MPI selector to select MPI implementation ABC this default selection will not take effect until you start a new shell e g logout and login again Other packages such as environment mod ules provide functionality that allows changing your environment to point to a new MPI imple mentation in the current shell The MPI selector was not meant to duplicate or replace that functionality The MPI selector functionali
68. 2 2 OxOfff reset only error counters of port 2 Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 140 perfquery R 32 2 Oxf000 reset only non error counters of port 2 1 Read local port s performance counters 7 perfquery Port counters Lid 6 port 1 1 CounterSelect 0x1000 SymbolkErrors 0 LinkRecovers 0 LinkDowned 0 0 RcvRemotePhysErrors 0 RcevSwRelayErrors 0 XmtDiscards 0 XmtConstraintErrors 0 RevConstraintErrors 0 LinkIntegrityErrors 0 ExcBufOverrunErrors 0 VLI15Dropped 0 55178210 RcviData 55174680 766366 RcvBktis ne 766315 2 Read performance counters from LID 2 all ports gt smpquery a 2 Port counters Lid 2 port 255 Porse 255 CounterSelect 0x0100 SymbolErrors 65535 LinkRecovers 255 LinkDowned 16 m 657 RcvRemotePhysErrors 0 RcvSwRelayErrors 70 XmtDiscards 488 XmtConstraintErrors 0 RevConstraintErrors 0 LinkIntegrity Errors 0 ExcBufOverrunErrors 0 VLI15Dropped 0
69. 26418 CONNECTX IB 26418 ROM X X XXX rom HCA Single Dual port ConnectX IB QDR amp PCI Express 2 0 5 0GT s PCI DevID 26428 CONNECTX IB 26428 ROM X X XXX rom HCA InfiniHost III Ex in Mem Free mode PCI DevID 25218 IHOST3EX PORT1 ROM X X XXX rom IB Port 1 IHOST3EX PORT2 ROM X X XXX rom IB Port 2 HCA InfiniHost III Lx PCI DevID 25204 IHOST3LX single IB Port device 2 Additional documents under docs dhcpd conf sample DHCP configuration file dhcp patch patch file for DHCP v3 1 2 Burning the Expansion ROM Image The binary code resides in the same Flash device of the device firmware Note that the binary files are distinct and do not affect each other Mellanox s mlxburn tool is available for burning how ever it is not possible to burn the expansion ROM image by itself Rather both the firmware and expansion ROM images must be burnt simultaneously mlxburn requires the following items 1 MST device name After installing the MFT package run mst start mst status The device name will be of the form dev mst mt dev id pci cr0 con f0 2 The firmware mlx file fw lt ID gt X X XXX mlx 3 One of the expansion ROM binary files listed in Section A 1 3 Firmware burning example The following command burns a firmware image and an expansion ROM image to the Flash device of a ConnectX adapter card mlxburn dev dev mst mt25
70. 4 IO class 0100 ID LSI Storage Systems SRP Driver 200400 0681146 1 service entries 1 service 0 200400a0b81146al SRP T10 200400A0B81146A1 b detect all the SRP Targets reachable by the SRP Initiator via another umad device use the following com mand ibsrpdm d umad device 2 Assistance in creating an SRP connection a To generate output suitable for utilization in the echo command of Section 7 2 2 add the option to ibsrpdm ibsrpdm c Sample output id_ext 200400A0B81146A1 ioc_guid 0002c90200402bd4 dgid fe800000000000000002c90200402bd5 pkey ffff service_id 200400a0b8 1146a1 b establish a connection with an SRP Target using the output from the libsrpdm c example above execute the following command echo n id ext 200400A0B81146A1 ioc guid 0002c90200402bqd4 dgid fe800000000000000002c90200402bd5 pkey ffff service id 200400a0b81146a1 gt sys class infiniband srp srp mthca0 1 add target The SRP connection should now be up the newly created SCSI devices should appear in the listing obtained from the disk 1 command srp daemon The srp daemon utility is based on ibsrpdm and extends its functionality In addition to the ibsrpdm functionality described above srp daemon can also Establish an SRP connection by itself without the need to issue the echo command described in Section 7 2 2 Continue running in background detecting new targets and establishing SRP conn
71. 418 pci cr0 fw fw 25408 X X XXX mlx V conf MHGH28 XTC ini exp rom ConnectX IB 25418 ROM X X XXX rom A 3 Preparing the DHCP Server in Linux Environment The DHCP server plays a major role in the boot process by assigning IP addresses for BoIB clients and instructing the clients where to boot from BoIB requires that the DHCP server runs on a machine which supports IP over IB Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 151 A 3 1 Configuring the DHCP Server A 3 1 1 For ConnectX Family Devices When a BoIB client boots it sends the DHCP server various information including its DHCP cli ent identifier This identifier is used to distinguish between the various DHCP sessions The value of the client identifier is composed of an 8 byte port GUID separated by colons and represented in hexadecimal digits Extracting the Port GUID Method I To obtain the port GUID run the following commands Note The following MFT commands assume that the Mellanox Firmware Tools MFT package has been installed on the client machine 1 mst start hostl mst status The device name will be of the form dev mst mt lt dev_id gt _pci _cr0 conf0 Use this device name to obtain the Port GUID via the following query command flint d lt MST DEVICE NAME gt q Example with ConnectX IB DDR amp PCI Express 2 0 2 5GT s as the HCA device hostl flint d dev mst mt25418 pci cr0 q Imag
72. 5 LinkUp rate 10 Gb sec 4X Infiniband device mlx4_0 port 2 status default gid fe80 0000 0000 0000 0000 0000 0007 3897 base lid 0x1 sm lid 0x1 state 4 ACTIVE phys state 5 LinkUp rate 20 Gb sec 4X DDR 11 8 ibportstate Applicable Hardware All InfiniBand devices Description Enables querying the logical link and physical port states of an InfiniBand port It also allows adjusting the link speed that is enabled on any InfiniBand port If the queried port is a swich port then ibportstate can be used to e disable enable or reset the port validate the port s link width and speed against the peer port Synopsis ibportstate d e v V D G s lt smlid gt C ca name P ca port t timeout ms lt dest dr path lid guid lt portnum gt lt op gt lt value gt Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 127 Table 28 lists the various flags of the command Table 28 ibportstate Flags and Options Fla Optional Default Description 3 Mandatory If Not Specified P h help Optional Print the help menu d ebug Optional Raise the IB debug level May be used several times for higher debug levels ddd or d d d e rr show Optional Show send and receive errors timeouts and others v erbose Optional Increase verbosity level May be used several times for addit
73. 8 SRP 86 9 9 OpenSM Features 86 Chapter 10 OpenSM Subnet Manager 87 10 1 Overview 87 10 2 opensm Description 87 10 2 1 Syntax 87 10 2 2 Environment Variables 93 10 2 3 Signaling 93 10 2 4 Running opensm 93 10 2 4 1 Running OpenSM As Daemon 93 10 3 osmtest Description 94 10 3 1 Syntax 94 10 3 2 Running osmtest 96 10 4 Partitions 97 10 4 1 File Format 97 10 5 Routing Algorithms 98 10 5 1 Effect of Topology Changes 99 10 5 2 Min Hop Algorithm 100 10 5 3 Purpose of UPDN Algorithm 100 10 5 3 1 Algorithm Usage 101 10 5 4 Fat tree Routing Algorithm 101 Mellanox Technologies Rev 1 10 6 Chapter 11 10 5 5 LASH Routing Algorithm 10 5 6 DOR Routing Algorithm 10 5 7 Routing References 10 5 8 Modular Routine Engine 10 6 Quality of Service Management in OpenSM 10 6 1 Overview 10 6 2 Advanced QoS Policy File 10 6 3 Simple QoS Policy Definition 10 6 4 Policy File Syntax Guidelines 10 6 5 Examples of Advanced Policy File 10 6 6 Simple QoS Policy Details and Examples 10 6 6 1 IPoIB 10 6 6 2 SDP 10 6 6 3 RDS 10 6 6 4 iSER 10 6 6 5 SRP 10 6 6 6 MPI 10 6 7 SL2VL Mapping and VL Arbitration 10 6 8 Deployment Example 10 7 QoS Configuration Examples 10 7 1 Typical HPC Example MPI and Lustre 10 7 2 EDC SOA 2 tier IPoIB and SRP 10 7 3 EDC 3 tier IPoIB RDS SRP InfiniBand Fabric Diagnostic Utilities 11 1 Overview 11 2 Utilities Usage 11 2 1 Common Configuration Interface and Addressing 11 2 2 IB Interface Def
74. 9 00 00 bb on PCIO2 00 0 open Link up TX 0 0 RX 0 0 Waiting for link up on 0 ok Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual B 7 B 7 1 179 If the Ethernet link comes up successfully the client attempts connecting to the DHCP server to obtain an IP address and the source location of the kernel OS to boot from The client waits up to 30 seconds for a DHCP server response Mellanox ConnectX Boot over IB v2 0 000 gPXE 0 9 6 Open Source Boot Firmware 0 00 02 9 00 01 77 70 51 on PCIO2 00 0 open Link down TX 0 TXE O RX 0 RXE 0 Waiting for link up on netO0 HCP netO 00 02 c9 00 01 77 70 51 ok 1 0 11 4 3 130 255 255 255 0 gw 0 0 0 0 Next ConnectX EN PXE attempts to boot as directed by the DHCP server Diskless Machines Mellanox ConnectX EN PXE supports booting diskless machines To enable using an Ethernet driver the remote kernel or initrd image must include and be configured to load the driver This can be achieved either by compiling the adapter driver into the kernel or by adding the device driver module into the initrd image and loading it The Ethernet driver requires loading the following modules in the specified order see Section B 7 1 for an example mlx4 core ko mlx4 en ko Example Adding an Ethernet Driver to initrd Linux Prerequisites 1 The ConnectX EN PXE image is already programmed
75. 9 16 8 5 4 Intel MPI Benchmark To run the Intel MPI Benchmark test enter host1 usr mpi gcc mvapich mvapich ver bin mpirun rsh np 2 hostfile home lt username gt cluster usr mpi gcc mvapich lt mvapich ver gt tests IMB lt IMB ver gt IMB MPI1 Hh ate ae Intel R MPI Benchmark Suite V3 0 MPI 1 part ee A A EELE ae Date Sun Mar 2 19 56 42 2008 Machine x86 64 System Linux Release 2 6 16 21 0 8 smp Version 1 SMP Mon Jul 3 18 25 39 UTC 2006 MPI Version a te MPI Thread Environment MPI THREAD FUNNELED H Minimum message length in bytes 0 Maximum message length in bytes 4194304 MPI _ Datatype MPI BYTE MPI Datatype for reductions MPI FLOAT MPI Op MPI SUM List of Benchmarks to run PingPong PingPing Sendrecv Exchange Allreduce Reduce Reduce_scatter Allgather Mellanox Technologies Rev 1 10 78 MPI Allgatherv Alltoall Alltoallv Beast Barrier Benchmarking PingPong processes 2 t usec Mbytes sec 0 00 0 77 1 52 3 09 6 07 11 83 22 51 40 25 45 74 80 48 134 22 199 69 309 85 438 24 555 20 828 93 1025 75 1164 57 1252 80 1303 44 1337 19 1347 14 1355 75 bytes repetitions 0 1000 1 25 1 1000 1 24 2 1000 1 25 4 1000 1 23 8 1000 1 26 16 1000 1 29 32 1000 1 36 64 1000 1 52 128 1
76. AF INET to 11 4 17 6 11 4 17 6 port 0 AF INET Local Remote Socket Size Request Resp Elapsed Trans CPU CPU S dem S dem Send Recv Size Size Time Rate local remote local remote bytes bytes bytes bytes secs per sec S 58 us Tr us Tr 16384 87380 1 1 10 00 37572 8983 15 172 23 36 33 469 49 729 16384 87380 The following table describes parameters for the netperf command Option Description H Where to find the server 11 4 17 6 SDP IPoIB IP address t Test Name Specify the test to perform Options are TCP STREAM TCP RR etc C Client CPU utilization C Server CPU utilization Separates the global and test specific parameters r 1 1 The request size sent and how many bytes requested back Note that the run example above produced the following results Mellanox Technologies Rev 1 10 SDP 64 e Client CPU utilization is 15 72 percent of client CPU Server CPU utilization is 23 36 percent of server CPU Latency is 13 31 microseconds Latency is calculated as follows 0 5 1 Transaction rate per sec 1 000 000 one way average latency in usec Step 7 To end the test shut down the Netperf server host1 pkill netserver Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 7 7 1 7 2 7 2 1 7 22 65 SRP Overview As described in Section 1 4 5 the SCSI RDMA Protocol SRP is designed to take
77. Disk Browse boot initrd v Browse Initial RAM Disk if Root Device not empty definesthe initial ramdisk to use Idev sda2 v Either enter the path and file name directly Mode or choose by using 0x332 Browse Root Device setsthe device to passto the a kernel as root device 71 Back Abort OK Step 17 If you wish to change additional settings click the appropriate item and perform the changes and click Accept when done Step 18 In the Confirm Installation window click Install to start the installation See image below Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 193 Preparation EEA 8 Installation Settings v License Agreement v Disk Activation Click headli KE CH he Ch bel v Xsystem nalysis ick any headline to make changes or use the Change menu below v Time Zone Overview Expert Installation Confirm Installation Configuration Hostname Root Password Network Customer Center Online Update All information required forthe base installation is now complete e Service If you continue now partitions on your hard disk will be formatted erasing any existing data in those partitions according to the installation settings in the previous dialogs Go back and check the settings if you are unsure Users Clean Up Release Notes Hardware Conf
78. Ds in range including invalid entries n o dests Optional Do not try to resolve destinations D irect Optional Use directed path address arguments The path is a comma sepa rated list of out ports Examples 0 self port 0 1 2 1 4 out via port 1 then 2 G uid Optional Use GUID address argument In most cases it is the Port GUID Example 0x08f1040023 M ulticast Optional Show multicast forwarding tables The parameters lt startlid gt and lt endlid gt specify the MLID range s lt smlid gt Optional Use lt smlid gt as the target LID for SM SA queries C lt ca_name gt Optional Use the specified channel adapter or router P lt ca_port gt Optional Use the specified port t lt timeout_ms gt Optional Override the default timeout for the solicited MADs msec Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 132 Table 29 ibportstate Flags and Options Fla Optional Default Deseripti n 8 Mandatory If Not Specified dest dr path lid Optional Destination s directed path LID or GUID guid lt startlid gt Optional Starting LID in an MLID range lt endlid gt Optional Ending LID in an MLID range Examples 1 Dump all Lids with valid out ports of the switch with Lid 2 gt ibroute 2 Unicast lids 0x0 0x8 of switch Lid 2 guid 0x0002c902fffff00a MT47396 Infiniscale III Mel lanox Technologies Lid Out
79. IS HARDWARE EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual Table of Contents Table of Contents List of Tables Revision History 11 Preface 12 Intended Audience 12 Documentation Conventions 12 Typographical Conventions 12 Common Abbreviations and Acronyms 13 Related Documentation 14 14 Chapter 1 Mellanox BXOFED Overview 15 1 1 Introduction to Mellanox BXOFED 15 1 2 Introduction to Mellanox VPI Adapters 15 1 3 BXOFED Package 15 1 3 1 Download Tarball 15 1 3 2 Software Components 16 1 4 Architecture 16 1 4 1 mthca HCA IB Driver 17 1 4 2 mlx4 VPI Driver 17 1 4 3 Mid layer Core 18 1 4 4 Open FCoE 18 1 4 5 ULPs 18 1 46 MPI 19 1 4 7 InfiniBand Subnet Manager 19 1 4 8 Diagnostic Utilities 19 1 4 9 Performance Utilities 19 1 5 Quality of Service 19 Chapter 2 Working With VPI 21 2 1 Port Type Management 21 22 InfiniBand Driver 22 2 3 Ethernet Driver 22 2 3 1 Overview 22 2 3 2 Loading the Ethernet Driver 22 2 3 3 Unloading the Driver 22 2 3 4 Ethernet Driver Usage and Configuration 23 2 4 Fibre Channel over Ethernet 24 2 4 1 Overview 24 2 4 2 Installation 24 2 4 3 FCoE Basic Usage 25 2 4 3 1 FCoE Configuration 25 2 4 3 2 Starting FCoE Service 25 2 4 3 3 Stopping FCoE Service 26 2 4 4 FCoE Advanced Usage 26 2 4 4 1 Manual vHBA Control 26 2 4 4 2 Creating VHBAs That Use PFC 26 2 4 4 3 Creating VHBAs That Use Link Pause 27 Chapter 3 I
80. Initiator Discovery Portal Address Target Name 110 4 3 7 3260 1 ign 2007 08 7 3 4 10 isi Connected csiboot False Connect Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 163 Tip If no iSCSI target was recognized then either the target was not properly installed or no connection was found between the client and the iSCSI target Open a shell to ping the iSCSI target you can use CTRL ALT F2 and verify that the target is or is not accessible To return to the graphical installation screen press CTRL ALT F7 Step 5 The iSCSI Initiator Discovery window will now request authentication to access the iSCSI target Click Next to continue without authentication unless authentication is required Preparation Language License Agreement gt Disk Activation e System Analysis e Time Zone iSCSI Initiator Discovery Installation e Installation Summary e Perform Installation Configuration e Root Password Hostname e Network Customer Center Incoming Authentication Online Update Username Password Service e Users e Clean Up e Release Notes e Hardware Configuration No Authentication Outgoing Authentication Username Password Help Back Abort Step 6 The iSCSI Initiator Discovery window will show the iSCSI target that got connected to Note that the Connected column must indicate
81. Language License Agreement 055 Click headline to make changes or use the Change menu below System Analysis Time Zone Overview Expert a Installation Settings Installation gt Installation Summary Keyboard Layout w Perform Installation English US Configuration ONES e Root Password Partitioning Hostname Create swap partition dev sdal 502 0 MB e Network Create root partition dev sda2 7 5 GB with reiserfs e Customer Center e Online Update Add On Products e Service e Users No add on product selected for installation e Clean Up e Release Notes Software e Hardware Configuration SUSE Linux Enterprise Server 10 SP2 Server Base System 4 KDE Desktop Environment for Server C C Compiler and Tools X Window System Size of Packagesto Install 1 6 GB Booting Boot Loader Type GRUB Location dev sda2 boot Sections SUSE Linux Enterprise Server 10 SP2 default Floppy Failsafe SUSE Linux Enterprise Show Release Notes Server 10 SP2 CTUM Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual Step 15 Click Edit in the Boot Loader Settings window Section List From Other you can manually editthe boot loader configuration files clearthe current configuration and propose a new configuration start from scratch or reread the configuration saved on your disk If you have multipl
82. MC values 0 should only be used if the subnet topology actually provides multiple paths between ports i e multiple interconnects between switches Without 1 OpenSM defaults to LMC 0 which allows one path between any two ports priority priority value This option specifies the SMA s priority This will affect the handover cases where the master is chosen by priority and GUID Range is 0 default and lowest priority to 15 highest smkey SM Key value r R This option specifies the SMA s SM Key 64 bits This will affect SM authentication Note that OpenSM version 3 2 1 and below used 1 as the default value in a host byte order now it is fixed but you may need this option to inter operate with an old OpenSM running on a little endian machine reassign lids This option causes OpenSM to reassign LIDs to all end nodes Specifying r on a running subnet may disrupt subnet traffic Without r OpenSM attempts to preserve existing LID assignments resolving multiple use of same LID routing engine Routing engine names This option chooses routing engine s to use instead of Min Hop algorithm default Multiple routing engines can be Specified separated by commas so that specific ordering of routing algorithms will b tried if earlier routing engines fail Supported engines minhop updn file ftree lash dor Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack f
83. Mellanox TECHNOLOGIES Mellanox BXOFED Stack for Linux User s Manual Rev 1 10 Mellanox Technologies 2 O Copyright 2010 Mellanox Technologies Inc All Rights Reserved Mellanox BridgeX ConnectX InfiniBlast InfiniBridge InfiniHost InfiniRISC InfiniScale and InfiniPCI are regis tered trademarks of Mellanox Technologies Ltd FabricIT PhyX and Virtual Protocol Interconnect are trademarks of Mellanox Technologies Ltd Mellanox BXOFED Stack for Linux User s Manual Document Number 3253 Mellanox Technologies Inc 350 Oakmead Parkway Suite 100 Sunnyvale CA 94085 U S A www mellanox com Tel 408 970 3400 Fax 408 970 3403 Mellanox Technologies Ltd PO Box 586 Hermon Building Yokneam 20692 Israel Tel 972 4 909 7200 Fax 972 4 959 3245 NOTE THIS INFORMATION IS PROVIDED BY MELLANOX FOR INFORMATIONAL PURPOSES ONLY AND ANY EXPRESS OR IMPLIED WARRANTIES INCLUDING BUT NOT LIMITED TO THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED IN NO EVENT SHALL MELLANOX BE LIABLE FOR ANY DIRECT INDIRECT INCIDENTAL SPECIAL EXEM PLARY OR CONSEQUENTIAL DAMAGES INCLUDING BUT NOT LIMITED TO PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES LOSS OF USE DATA OR PROFITS OR BUSINESS INTERRUPTION HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY WHETHER IN CONTRACT STRICT LIABIL ITY OR TORT INCLUDING NEGLIGENCE OR OTHERWISE ARISING IN ANY WAY OUT OF THE USE OF TH
84. Notes ctc WIndawsystem 3 GNOME Desktop Environment for Server Hardware Configuration 4 Server Base System Novell AppArmor 4 Print Server Size of Packagesto Install 1 3 GB Language Primary Language English US Show Release Notes Help Abort Step 11 Select Base Partition Setup on This Proposal then click Next Your hard disks have Suggested Partitioning been checked The partition setup displayed is proposed for your hard drive Create boot partition dev sdal 70 5 MB with ext2 Create swap partition dev sda2 502 0 MB To accept these Create root partition dev sda3 7 4 GB with reiserfs suggestions and continue select Accept Proposal Ifthe suggestion does not fit your needs create your own partition setup starting with the partitions as currently present on the disks For this select Custom Partition Setup This is also the option to choose for advanced options like RAID and LVM Partitioning Accept Proposal Base Partition Setup on This Proposal Create Custom Partition Setup Back Abort Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 189 Step 12 In the Expert Partitioner window select from the IET VIRTUAL DISK device the row that has its Mount column indicating swap then click Delete Confirm the delete opera tion and click Finish
85. P server as described in Section A 3 Configure and start at least one of the services iSCSI Target see Section A 9 and or TFTP see Section A 5 7 2 Starting Boot Boot the client machine and enter BIOS setup to configure MLNX IB for ConnectX family or gPXE for InfiniHost III family to be the first on the boot device priority list see Section Note On dual port network adapters the client first attempts to boot from Port 1 If this fails it switches to boot from Port 2 Note also that the driver waits up to 45 sec for each port to come up Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 155 If MLNX IB gPXE was selected through BIOS setup the client will boot from BoIB The client will display BoIB attributes and wait for IB port configuration by the subnet manager For ConnectX Mellanox Ct X Boot over IB v2 0 000 gPXE 0 9 64 Open Source Boot Firmware net0 00 02 c9 00 01 77 70 51 on PCIO2 00 0 open Link down TX 0 TXE 0 RXE 0 link up on net0 ok 11 Boot over IB for InfiniHost III Ex oading via IB Port 2 for Infiniband link up ok After configuring the IB port the client attempts connecting to the DHCP server to obtain an IP address and the source location of the kernel OS to boot from For ConnectX Mellanox Ct X Boot over IB v2 0 000 gPXE 0 9 6 Open Source Boot Firmware 1etO 00 02 c9 00 01 77 70 51 on PCIO2 00
86. PR requests that didn t match any of the matching rules Any section subsection of the policy file is optional Examples of Advanced Policy File As mentioned earlier any section of the policy file is optional and the only mandatory part of the policy file is a default QoS Level Here s an example of the shortest policy file qos levels qos level name DEFAULT sl 0 end qos level end qos levels Port groups section is missing because no match rules which means that port groups not referred anywhere and there is no need defining them And since this policy file doesn t have any matching rules PR MPR query will not match any rule and OpenSM will enforce default QoS level Essentially the above example is equivalent to not having a QoS policy file at all The following example shows all the possible options and keywords in the policy file and their syn tax See the comments in the following example They explain different keywords and their meaning port groups port group using port GUIDs name Storage use is just a description that is used for logging Other than that it is just a comment use SRP Targets port guid 0x10000000000001 0x10000000000005 0x1000000000FFFA 107 Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 108 port guid 0 1000000000 end port group port group name Virtual Servers The syntax of the port name is
87. PoIB 29 3 1 Introduction 29 Mellanox Technologies Rev 1 10 4 3 2 IPoIB Mode Setting 29 3 3 IPoIB Configuration 29 3 3 1 IPoIB Configuration Based on DHCP 30 3 3 1 1 DHCP Server 30 3 3 1 2 DHCP Client Optional 30 3 3 2 Static IPoIB Configuration 31 3 3 3 Manually Configuring IPoIB 32 3 4 Subinterfaces 32 3 4 1 Creating a Subinterface 32 3 4 2 Removing a Subinterface 33 3 5 Verifying IPoIB Functionality 33 3 6 The ib bonding Driver 34 3 6 1 Using the ib bonding Driver 34 3 7 IPoIB Performance Tuning 35 3 8 Testing IPoIB Performance 35 Chapter 4 EoIB 39 4 1 Introduction 39 42 EoIB Topology 39 4 2 1 External ports eports and GW 40 4 2 2 Virtual Hubs vHubs 40 4 2 3 Virtual NIC vNic 40 4 3 EoIB Configuration 41 4 3 1 EoIB Host Administered vNic 41 4 3 1 1 Central configuration file etc infiniband mlx4 vnic conf 42 4 3 1 2 vNic specific configuration files ifcfg ethX 42 4 3 2 Extracting BridgeX host name 43 4 3 3 mlx4 confd 43 4 3 4 EoIB Network Administered vNic 44 4 3 5 VLAN Configuration 44 4 3 5 1 Configuring VLANs 45 4 3 6 EoIB Multicast Configuration 45 4 3 7 EoIB and QoS 45 4 3 8 IP Configuration Based on DHCP 45 4 3 8 1 DHCP Server 46 4 3 9 Static EoIB Configuration 46 4 4 Sub Interfaces VLAN 46 4 5 Retrieving EoIB information 46 4 5 1 mlx4 vnic info 46 4 5 2 ethtool 48 4 5 3 Link state 48 4 6 Bonding Driver 49 4 7 Jumbo Frames 49 4 8 Module Parameter
88. VICE The name of the interface that is displayed when running ifconfig This field is optional if it is not present the trailer of the configuration file name e g ifcfg eth47 gt eth47 is used BXADDR The BridgeX box system GUID or system name string BXEPORT The string describing the eport name VNICVLAN An optional field If it exists the vNic will be assigned the VLAN id specified This value must be between 0 and 4095 VNICIBPORT Device name and port number in the form device name port number The device name can b retrieved by running ibv devinfo and using the out put of hca id filed The port number can have a value of 7 3052 HWADDR The mac address to assign the vnic Other fields available for regular ethernet interfaces in the ifcfg ethX files may also be used Extracting BridgeX host name In order to configure host administered vNics the BridgeX box address needs to be known on the host The simplest way to learn this information is by loging into the BridgeX box and querying the information One way to perform this is by connecting through ssh to the BridgeX and running bridge 1128b8 4 enable bridge 1128b8 configure terminal bridge 1128b8 config show bxm The result will display the system GUID and some more information BXM status System GUID 00 02 C9 03 00 11 08 C7 An alternative 1s to replace the last command with bridge 1128b8 config show hosts The result will displ
89. a0 root lab104 ibsrpdm c d dev infiniband umad0 id_ext 0002c90200226cf4 i0c_guid 0002c90200226cf4 dgid fe800000000000000002c90200226cf5 pkey ffff service_id 0002c90200226cf4 root lab104 echo id_ext 0002c90200226cf4 i0c_guid 0002c90200226cf4 dgid fe800000000000000002c90200226cf5 pkey ffffservice_id 0002c90200226cf4 gt sys class infiniband_srp srp mthca0 1 add_target OR You can edit etc infiniband openib conf to load srp driver and srp HA daemon automatically that is set SRP_LOAD yes and SRPHA_ENABLE yes To set up and use high availability feature you need dm multipath driver and multipath tool Please refer to OFED 1 x SRP s user manual for more in details instructions on how to enable use the HA feature The following is an example of ab SRP Target setup file bin sh modprobe scst scst_threads 1 modprobe scst_vdisk scst_vdisk_ID 100 echo open vdisk0 dev cciss c1d0 BLOCKIO gt proc scsi_tgt vdisk vdisk echo open vdisk1 dev sdb BLOCKIO gt proc scsi_tgt vdisk vdisk echo open vdisk2 dev sde BLOCKIO gt proc scsi_tgt vdisk vdisk echo open vdisk3 dev sdd BLOCKIO gt proc scsi_tgt vdisk vdisk echo add vdisk0 0 gt proc scsi_tgt groups Default devices echo add vdisk1 1 gt proc scsi tgt groups Default devices echo add vdisk2 2 gt proc scsi_tgt groups Default devices echo add vdisk3 3 gt proc scsi tgt groups Default devices modprobe ib srpt echo add mgmt gt
90. able If you do not have it in your initrd please add it using the following command host1 cp sbin insmod tmp initrd en sbin Step 6 If you plan to give your Ethernet device a static IP address then copy ifconfig Other wise skip this step host1 cp sbin ifconfig tmp initrd en sbin Step 7 Now you can add the commands for loading the copied modules into the file init Edit the file tmp initrd en init and add the following lines at the point you wish the Ethernet driver to be loaded Warning The order of the following commands for loading modules is critical echo loading Mellanox ConnectX EN driver sbin insmod lib modules mlnx en mlx4 core ko sbin insmod lib modules mlnx en mlx4 en ko Step 8 Now you can assign a static or dynamic IP address to your Mellanox ConnectX EN net work interface Step 9 Savethe init file Step 10 Close initrd host1 cd tmp initrd en hostl find cpio H newc o gt tmp new initrd en img hostl gzip tmp new init en img Step 11 At this stage the modified initrd including the Ethernet driver is ready and located at tmp new init ib img gz Copy it to the original initrd location and rename it properly iSCSI Boot Mellanox ConnectX EN PXE enables an 1SCSI boot of an OS located on a remote iSCSI Target It has a built in iSCSI Initiator which can connect to the remote iSCSI Target and load from it the ker nel and initrd There are two instances of connection to the rem
91. ace To create a child interface subinterface follow this procedure Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 3 4 2 3 5 33 Note In the following procedure ib0 is used an example of an IB subinterface Step 1 Decide on the PKey to be used in the subnet Valid values are 0 255 The actual PKey used is a 16 bit number with the most significant bit set For example a value of 0 will give a PKey with the value 0x8000 Step 2 Create a child interface by running host1 echo lt PKey gt gt sys class net lt IB subinterface gt create child Example host1 echo 0 gt sys class net ib0 create_child This will create the interface ib0 8000 Step 3 Verify the configuration of this interface by running host1 ifconfig lt subinterface gt lt subinterface PKey gt Using the example of Step 2 host1 ifconfig ib0 8000 ib0 8000 Link encap UNSPEC HWaddr 80 00 00 4A FE 80 00 00 00 00 00 00 00 00 00 00 BROADCAST MULTICAST MTU 2044 Metric 1 RX packets 0 errors 0 dropped 0 overruns 0 frame 0 TX packets 0 errors 0 dropped 0 overruns 0 carrier 0 collisions 0 txqueuelen 128 RX bytes 0 0 0 b TX bytes 0 0 0 b Step 4 As can be seen the interface does not have IP or network addresses To configure those you should follow the manual configuration procedure described in Section 3 3 3 Step 5 To be able to use this interface a configuration of the Subnet Manager is needed
92. ace SDP debug is controlled by options in the Libsdp conf file You can also have a local version and point to it explicitly using the following command Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 55 hostl export LIBSDP CONFIG FILE lt path gt libsdp conf To obtain extensive debug information you can modify 1ibsdp conf to have the log directive produce maximum debug output provide the min level flag with the value 1 The 1og statement enables the user to specify the debug and error messages that are to be sent and their destination The syntax of 1og is as follows log destination stderr syslog file lt filename gt min level 1 9 where options are destination send log messages to the specified destination stderr forward messages to the STDERR syslog send messages to the syslog service file lt filename gt write messages to the file var log filename for root For a regular user write to tmp lt filename gt lt uid gt if filename is not specified as a full path otherwise write to lt path gt lt filename gt lt uid gt min level verbosity level of the log 9 print errors only 8 print warnings 7 print connect and listen summary useful for tracking SDP usage 4 print positive match summary useful for config file debug 3 print negative match summary useful for config file debug 2 print function calls and return values 1 pr
93. artition configuration file The default name is etc opensm partitions conf prefix routes file file name Prefix routes control how the SA responds to path record que ries for off subnet DGIDs By default the SA fails such queries The PREFIX ROUTES section below describes the for mat of the configuration file The default path is etc opensm prefix routes conf Q qos This option enables QoS setup It is disabled by default Y qos policy file file name gt This option defines the optional QoS policy file The default name is etc opensm qos policy conf N no part enforce This option disables partition enforcement on switch external ports y stay on fatal This option will cause SM not to exit on fatal initializa tion issues if SM discovers duplicated guids or a 12x link with lane reversal badly configured By default the SM will exit on these errors B daemon Run in daemon mode OpenSM will run in the background I inactive Start SM in inactive rather than init SM state This option can be used in conjunction with the perfmgr so as to run a standalone performance manager without SM SA However this is NOT currently implemented in the performance manager perfmgr Enable the perfmgr Only takes effect if enable perfmgr was specified at configure time Mellanox Technologies Rev 1 10 OpenSM Subnet Manager
94. ation test RMPP stress test Small SA Queries stress test 10 3 1 Syntax osmtest OPTIONS where OPTIONS are f flow This option directs osmtest to run a specific flow Flow Description C create an inventory file with all nodes ports and paths a run all validation tests expecting an input inventory v only validate the given inventory file S run service registration deregistration and lease test e run event forwarding test f flood the SA with queries according to the stress mode m multicast flow q QoS info dump VLArb and SLtoVL tables t run trap 64 65 flow this flow requires running of exter nal tool Default all flows except QoS w wait This option specifies the wait time for trap 64 65 in sec onds It is used only when running f t the trap 64 65 flow Default 10 sec d debug This option specifies a debug option These options are not normally needed The number following d selects the debug option to enable as follows OPT Description q0 Ignore other SM nodes di Force single threaded dispatching d2 Force log flushing after each log message d3 Disable multicast support Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual m g Pr i s M t EU max lid guid port inventory Stress This option specifies the maximal LID number to be searched for during inventory file build Default 100 This option s
95. ay the system name and some more information Hostname bridge 1128b8 mlx4 vnic confd After updating the configuration files you are ready to create the host administered vNics To create the vNics you will use the mlx4 vnic confd service Mellanox Technologies Rev 1 10 44 EoIB 4 3 4 4 3 5 that is located at etc init d To use the service run To start load new vnics mlx4 vnic confd start To stop all host administrated vNics mlx4 vnic confd stop To restart close and then open all host administrated vNics mlx4 vnic confd restart To update system according to up to date configuration files This option will not modify vNics with a valid configuration that was not changed mlx4 vnic confd reload EoIB Network Administered vNic In network administered mode the configuration of the vNic is done by the BridgeX If a vNic is configured for a specific host it will appear on that host once a connection is established between the BridgeX and the mlx4 vnic module This connection between the mlx4 vnic modules and all available BridgeX boxes is established automatically when the mlx4 vnic module is loaded If the BridgeX is configured to remove the vnic or if the connection between the host and BridgeX is lost the vNic interface will disappear running ifconfig will not display the interface Similar to host administered vNics a network administered vNic resides on a specific vHub See BridgeX docu
96. b port vid vnic id bx eport The name of the interface that is displayed when running ifconfig The mac address to assign to the vnic Device name and port number in the form device name port number The device name can b retrieved by running ibv devinfo and using the out put of hca id field The port number can have a value of 1 or 2 vLan ID an optional field If it exists the vNic will be assigned the VLAN id specified This value must be between 0 and 4095 If no vid is specified the vNic will be assigned to the default vHub asso ciated with the bx eport pair GW A unique number per vNic between 0 and 32K The BridgeX box system GUID or system name string The string describing the eport name 4 3 1 2 vNic specific configuration files ifefg ethX EoIB configuration can use the well known ifcfg ethX files used by the network service to derive the needed configuration In this usage model a separate file is required per vNic We will need to update the ifcfg ethX file and add some new attributes to it On Red Hat the new file will be of the form DEVICE eth2 HWADDR 00 30 48 7d de e4 BOOTPROTO dhcp ONBOOT yes BXADDR BX001 BXEPORT A10 VNICVLAN 0 VNICIBPORT m1x4_0 1 Optional field Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 4 3 2 4 3 3 43 The fields used in the file have the following meaning DE
97. both ends of the cable In the case of meshes the dimension should consistently use the same pair of ports one port on one end of the cable and the other port on the other end continuing along the mesh dimension Use R dor option to activate the DOR algorithm Routing References To learn more about deadlock free routing see the article Deadlock Free Message Routing in Multiprocessor Interconnection Networks by William J Dally and Charles L Seitz 1985 To learn more about the up down algorithm see the article Effective Strategy to Compute For warding Tables for InfiniBand Networks by Jose Carlos Sancho Antonio Robles and Jose Duato at the Universidad Politecnica de Valencia To learn more about LASH and the flexibility behind it the requirement for layers performance comparisons to other algorithms see the following articles Layered Routing in Irregular Networks Lysne et al IEEE Transactions on Parallel and Dis tributed Systems VOL 16 No12 December 2005 103 Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 104 Routing for the ASI Fabric Manager Solheim et al IEEE Communications Magazine Vol 44 No 7 July 2006 e Layered Shortest Path LASH Routing in Irregular System Area Networks Skeie et al IEEE Computer Society Communication Architecture for Clusters 2002 10 5 8 Modular Routine Engine Modular routing engine structure allows for the ease of plugging
98. c infiniband openib conf and setting SET IPOIB After changing the mode you need to restart the driver by running etc init d openibd restart To check the current mode used for out going connections enter cat sys class net ib n mode IPoIB Configuration Unless you have run the installation script install with the flag n then IPoIB has not been configured by the installation The configuration of IPoIB requires assigning an IP address and a subnet mask to each HCA port like any other network adapter card 1 you need to prepare a file called ifcfg ib n for each port The first port on the first HCA in the host is called interface 160 the second port is called 161 and so on An IPoIB configuration can be based on DHCP Section 3 3 1 or on a static configuration Section 3 3 2 that you need to supply You can also apply a manual configuration that persists only until the next reboot or driver restart Section 3 3 3 Mellanox Technologies Rev 1 10 IPoIB 30 3 3 1 IPoIB Configuration Based on DHCP Setting an IPoIB interface configuration based on DHCP v3 1 2 which is availabe via www isc org is performed similarly to the configuration of Ethernet interfaces In other words you need to make sure that IPoIB configuration files include the following line For RedHat BOOTPROTO dhcp For SLES BOOTPROTO dchp 3 3 1 1 3 3 1 2 Note If IPoIB configuration files are in
99. chnologies Mellanox BXOFED Stack for Linux User s Manual 209 Glossary The following is a list of concepts and terms related to InfiniBand in general and to Subnet Manag ers in particular It is included here for ease of reference but the main reference remains the Infini Band Architecture Specification Channel Adapter CA Host Channel Adapter HCA An IB device that terminates an IB link and executes transport functions This may be an HCA Host CA or a TCA Target CA HCA Card A network adapter card based on an InfiniBand channel adapter device IB Devices Integrated circuit implementing InfiniBand compliant communication IB Cluster Fabric Subnet A set of IB devices connected by IB cables In Band A term assigned to administration activities traversing the IB connectivity only LID An address assigned to a port data sink or source point by the Subnet Manager unique within the sub net used for directing packets within the subnet Local Device Node System The IB Host Channel Adapter HCA Card installed on the machine running IBDIAG tools Local Port The IB port of the HCA through which IBDIAG tools connect to the IB fabric Master Subnet Manager The Subnet Manager that is authoritative that has the reference configuration information for the subnet See Subnet Manager Multicast Forwarding Tables A table that exists in every switch providing the list of ports to forward received m
100. cluded ifcfg ib lt n gt files will be installed under etc sysconfig network scripts ona RedHat machine etc sysconfig network ona SuSE machine Note A patch for DHCP is required for supporting IPoIB The patch file for DHCP v3 1 2 dhcp patch is available under the docs directory Standard DHCP fields holding MAC addresses are not large enough to contain an IPoIB hardware address To overcome this problem DHCP over InfiniBand messages convey a client identifier field used to identify the DHCP session This client identifier field can be used to associate an IP address with a client identifier value such that the DHCP server will grant the same IP address to any client that conveys this client identifier Note Refer to the DHCP documentation for more details how to make this association The length of the client identifier field is not fixed in the specification For BXOFED it is recom mended to have IPoIB use the same format that Boot over IB uses for this client identifier see Section A 3 1 Configuring the DHCP Server on page 151 DHCP Server In order for the DHCP server to provide configuration records for clients an appropriate configura tion file needs to be created By default the DHCP server looks for a configuration file called dhcpd conf under etc You can either edit this file or create a new one and provide its full path to the DHCP server using the cf flag See a file example at docs dhcpd conf of this package
101. console off local socket loopback This option brings up the OpenSM console default off Note that the socket and loopback options will only be available if OpenSM was built with enable console socket console port lt port gt Specify an alternate telnet port for the socket console default 10000 Note that this option only appears if OpenSM was built with enable console socket i ignore guids lt equalize ignore guids file gt This option provides the means to define a set of ports by node guid and port number that will be ignored by the link load equalization algorithm honor guid2lid This option forces OpenSM to honor the guid2lid file when it comes out of Standby state if such file exists under OSM CACHE DIR and is valid By default this is FALSE f log file file name This option defines the log to be the given file By default the log goes to var log opensm log For the log to go to standard output use f stdout Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual e log limit size in MB This option defines maximal log file size in MB When spec ified the log file will be truncated upon reaching this limit erase log file This option will cause deletion of the log file if it previously exists By default the log file is accumula tive Pconfig partition config file This option defines the optional p
102. cription Node Portl Port2 Sys image GUIDs 0002c90200231390 0002c90200231391 0002c90200231392 0002c90200231393 Board ID MT 0370110001 VSD PSID MT 0370110001 Assuming that BoIB is connected via Port 2 then the Port GUID is 00 02 c9 02 00 23 13 92 Step 4 The resulting client identifier is the concatenation from left to right of 20 the QP Number the subnet prefix and the Port GUID In the example above this yields the following DHCP client identifier 20 00 55 04 01 fe 80 00 00 00 00 00 00 00 02 c9 02 00 23 13 92 Extracting the Client Identifier Method II An alternative method for obtaining the 20 bytes of QP Number and GID involves booting the cli ent machine via BoIB This requires having a Subnet Manager running on one of the machines in the InfiniBand subnet The 20 bytes can be captured from the boot session as shown in the figure below IB for InfiniHost III Ex ver ort 2 for Infiniband link up ok gPXE 0 9 3 Open Sourc Featurgze PFTP ISCSI Aol 00550401 fe800000 00000000 0002c902 00231392 CI05 00 0 open TX 0 TXI LA netO 00550401 Ie rerexqererererererevETeTo 0 Pa 1 10 11 4 3 130 25 255 0 Concatenate the byte 20 to the left of the captured 20 bytes then separate every byte two hexa decimal digits with a colon You should obtain the same result shown in Step 4 above Placing Client Identifiers in etc dhcpd conf The following is an excerpt of a etc dhcpd conf exampl
103. d by your FCoE switch Note Only pre T11 format is offloaded in hardware DCBX IFS provide a space separated list of Ethernet devices to monitor the use of the DCBX protocol for the FCoE feature availability vHBAs are automatically created on these interfaces if the FCoE switch is configured for automatic FCoE negotiation MTU if MTU of the Ethernet device is changed from the default 1500 put the correct value here Configure the mlx4 en Ethernet driver to support PFC Add the following line to the file etc modprobe conf and restart the network driver options mlx4 en pfctx Oxff pferx Oxff 2 4 3 2 Starting FCoE Service Make sure the network is up modprobe mlx4 en Then run gt letc init d mlxfc start vHBAs will be instantiated on DCBX monitored interfaces and SCSI LUNs will get mapped For Manual instantiation of VHBAs please see Section 2 4 4 1 Manual vHBA Control Mellanox Technologies Rev 1 10 25 Working With VPI 26 2 4 3 3 Stopping FCoE Service Run gt etc init d mlxfe stop Note Only when the mlxfc service is stopped and the mlx4 en module is removed can the mlx4 core module be removed as well 2 4 4 FCoE Advanced Usage Advanced usage will probably be needed when connected to FCoE switches that do not support the Cisco like FCoE DCBX auto negotiation 2 4 4 1 Manual vHBA Control Manual control allows creating and destroying vHBAs and signaling link up and li
104. d core ib cm ko tmp initrd ib lib modules ib hostl cp infiniband core ib uverbs ko tmp initrd ib lib modules ib hostl cp infiniband core ib ucm ko tmp initrd ib lib modules ib hostl cp infiniband core ib umad ko tmp initrd ib lib modules ib hostl cp infiniband core iw cm ko tmp initrd ib lib modules ib hostl cp infiniband core rdma cm ko tmp initrd ib lib modules ib hostl cp infiniband core rdma ucm ko tmp initrd ib lib modules ib hostl cp net mlx4 mlx4 core ko tmp initrd ib lib modules ib hostl cp infiniband hw mlx4 mlx4 ib ko tmp initrd ib lib modules ib hostl cp infiniband hw mthca ib mthca ko tmp initrd ib lib modules ib hostl cp infiniband ulp ipoib ib ipoib ko tmp initrd ib lib modules ib IB requires loading an IPv6 module If you do not have it in your initrd please add it using the following command host1 cp lib modules uname r kernel net ipv6 ipv6 ko tmp initrd ib lib modules To load the modules you need the insmod executable If you do not have it in your initrd please add it using the following command host1 cp sbin insmod tmp initrd ib sbin If you plan to give your IB device a static IP address then copy ifconfig Otherwise skip this step host1 cp sbin ifconfig tmp initrd ib sbin If you plan to obtain an IP address for the IB device through DHCP then you need to copy the DHCP client which was compiled specifically to support IB Otherwise skip this step To continue with this step DHCP
105. direct route must be equal to the one specified in the option Oth erwise an error is reported Note When ibdiagpath queries for the performance counters along the path between the source and destination ports it always traverses the LID route even if a directed route is speci fied If along the LID route one or more links are not in the ACTIVE state ibdiagpath reports an error Moreover the tool allows omitting the source node in LID route addressing in which case the local port on the machine running the tool is assumed to be the source SYNOPSYS ibdiagpath n src name dst name 1 src lid dst lid d lt pl p2 p3 gt c lt count gt v t lt topo file gt s lt sys name gt ic lt dev index gt c p lt port num gt o lt out dir gt lw lt 1 4 12 gt ls lt 2 5 5 10 gt pm pc P lt lt PM counter gt lt Trash Limit gt gt OPTIONS n src name dst name Names of the source and destination ports as defined in the topology file source may be omitted gt local port is assumed to be the source 1 src lid dst lid Source and destination LIDs source may be omitted gt the local port is assumed to be the source d pl p2 p3 Directed route from the local node which is the source and the destination node c count The minimal number of packets to be sent across each link default 100 v Enable verbose m
106. disk I This command lists all devices the new devices are included in this listing Execute dmesg or look at var log messages to find messages with the names of the new devices 7 2 3 SRP Tools ibsrpdm and srp daemon To assist in performing the steps in Section 6 the BXOFED distribution provides two utilities ibsrpdm and srp daemon which Detect targets on the fabric reachable by the Initiator for Step 1 Output target attributes in a format suitable for use in the above echo command Step 2 The utilities can be found under usr sbin and are part of the srptools RPM that may be installed using the Mellanox BXOFED installation Detailed information regarding the various options for these utilities are provided by their man pages Below several usage scenarios for these utilities are presented ibsrpdm Ibsrpdm is using for the following tasks 1 Detecting reachable targets a detect all targets reachable by the SRP initiator via the default umad device dev umad0 execute the follow ing command ibsrpdm This command will output information on each SRP Target detected in human readable form Sample output IO Unit Info port LID 0103 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 67 port GID fe800000000000000002c90200402bd5 change ID 0002 max controllers 0x10 controller 1 GUID 0002c90200402bd4 vendor ID 0002c9 device ID 005a4
107. distinguish between the various DHCP sessions The value of the client identifier is composed of 21 bytes separated by colons having the follow ing components 20 lt QP Number 4 bytes gt lt GID 16 bytes Note Bytes are represented as two hexadecimal digits Extracting the Client Identifier Method I The following steps describe one method for extracting the client identifier 1 QP Number equals 00 55 04 01 for InfiniHost III Ex and InfiniHost III Lx HCAs 2 GID is composed of an 8 byte subnet prefix and an 8 byte Port GUID The subnet prefix is fixed for the supported Mellanox HCAs and is equal to fe 80 00 00 00 00 00 00 The next steps explains how to obtain the Port GUID To obtain the Port GUID run the following commands Note The following MFT commands assume that the Mellanox Firmware Tools MFT package has been installed on the client machine hostl mst start hostl mst status Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 153 The device name will be of the form dev mst mt lt dev_id gt _pci _cr0 conf0 Use this device name to obtain the Port GUID via a query command flint d lt MST DEVICE NAME gt q Example with InfiniHost III Ex as the HCA device hostl flint d dev mst mt25218 pci cr0 q Image type Failsafe FW Version Sad Rom Info type GPXE version 1 0 0 devid 25218 port 2 I S Version 1 Device ID 25218 Chip Revision AO Des
108. e For UDP best performance please use IPoIB in Datagram mode and not in Connected mode To verify IPoIB performance perform the following steps Step 1 Download Netperf from the following URL http www netperf org netperf NetperfPage html Step 2 Compile Netperf by following the instructions at http www netperf org netperf NetperfPage html Step 3 Start the Netperf server The following example shows how to start the Netperf server host1 netserver Starting netserver at port 12865 Starting netserver at hostname 0 0 0 0 port 12865 and family AF UNSPEC hostl Mellanox Technologies Rev 1 10 36 Step 4 Run the Netperf client The default test is the Bandwidth test The following example shows how to run the Netperf client which starts the Bandwidth test by default host2 netperf H 11 4 17 6 t TCP_STREAM c C m 65536 TCP STREAM TEST from 0 0 0 0 0 0 0 0 port 0 AF_INET to 11 4 17 6 11 4 17 6 port 0 AF_INET Recv Send Send Utilization Service Demand Socket Socket Message Elapsed Send Recv Send Recv Size Size Size Time Throughput local remote local remote bytes bytes bytes secs 10 6bits s S S us KB us KB 87380 16384 65536 10 00 2483 00 7 03 5 42 1 854 1 431 Note You must specify the IPoIB IP address when running the Netperf client The following table describes parameters for the netperf command Option Description H Where to find the server 11 4 17 6 IPoIB IP address
109. e R updn option instead of old u to activate the UPDN algorithm e Use a root guid file gt for adding guid file that contains the root nodes for ranking If the a option is not used OpenSM uses its auto detect root nodes algorithm Notes on the guid list file 1 A valid guid file specifies one guid in each line Lines with an invalid format will be discarded 2 The user should specify the root switch guids However it is also possible to specify CA guids OpenSM will use the guid of the switch if it exists that connects the CA to the subnet as a root node Fat tree Routing Algorithm The fat tree algorithm optimizes routing for shift communication pattern It should be chosen if a subnet is a symmetrical or almost symmetrical fat tree of various types It supports not just K ary N Trees by handling for non constant K cases where not all leafs CAs are present any CBB ratio As in UPDN fat tree also prevents credit loop deadlocks If the root guid file is not provided a root guid file options the topology has to be pure fat tree that complies with the following rules Tree rank should be between two and eight inclusively Switches of the same rank should have the same number of UP going port groups unless they are root switches in which case the shouldn t have UP going ports at all Switches of the same rank should have the same number of DOWN going port groups unless the
110. e Booting will oe skip all file system checks Optional Kernel r Section Settings Command Line Parameter lets you define additional Optional Kernel Command Line Parameter ponme jniphss to resume dev sdal splash silent showopts ibft mode off Kernel Image C Do not verify Filesystem before Booting Kernel Image defines the kernel to boot Iboot vmlinuz Browse Either enter the name li Sr es directly or choose via Initial RAM Disk Brows boot initrd Browse Initial RAM Disk if Root Device not empty defines the EJ ap initial ramdisk to use Idev sda2 r Either enterthe path and file name directly Mode or choose by using 0x332 Browse Root Device sets the device to passto the kernel as root device Back Abort OK Step 17 If you wish to change additional settings click the appropriate item and perform the changes and click Accept when done Step 18 In the Confirm Installation window click Install to start the installation See image below Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual Preparation v Language v License Agreement v Disk Activation v System Analysis v Time Zone 8 Installation Settings Click any headline to make changes or use the Change menu below Overview Expert Installation Confirm Installation Configuration
111. e Linux systems installed YaST can try to find them and merge their menus Boot Loader Settings Section Management Boot Loader Installation SUSE Linux Enterprise Server 10 Other end resume chainloader dev fdO Failsafe SUSE Linux Enterprise Server 10 SP2 Image append showopts ide n Ly Abort Set as Default 191 Mellanox Technologies Rev 1 10 192 Step 16 In the Optional Kernel Command Line Parameter field append the following string to the end of the line ibft mode off include a space before the string Click OK and then Finish to apply the change Section Name a Boot Loader Settings Section Management Use Section Name to specify the boot loader section name The section name must be unique Section Editor Section Settings Selecting Do not Section Name verify Filesystem suse Linux Enterprise Server 10 SP2 before Booting will v areas skip all file system checks Optional Kernel r Section Settings Command Line ae 7 Parameter lets you _ Do not verify Filesystem before Booting define additional Optional Kernel Command Line Parameter fra kamal alll ihe Kerne resume dev sdal splash silent showopts ibft mode off Kernel Image defines Ber eau ne the kernel to boot bootvmlinuz Browse Either enterthe name directly or choose via Initial RAM
112. e file showing the format of represent ing a client machine for the DHCP server Mellanox Technologies Rev 1 10 154 host host1 next server 11 4 3 7 filename pxelinux 0 fixed address 11 4 3 130 option dhcp client identifier V 20 00 55 04 01 fe 80 00 00 00 00 00 00 00 02 c9 02 00 23 13 92 j A 4 Subnet Manager OpenSM BoIB requires a Subnet Manager to be running on one of the machines in the IB network OpenSM is part of Mellanox BXOFED for Linux and can be used to accomplish this Note that OpenSM may be run on the same host running the DHCP server but it is not mandatory For details on OpenSM see OpenSM Subnet Manager on page 87 A 5 TFTP Server When you set the filename parameter in your DHCP configuration file to a non empty filename the client will ask for this file to be passed through TFTP For this reason you need to install a TFTP server A 6 BIOS Configuration The expansion ROM image presents itself to the BIOS as a boot device As a result the BIOS will add to the list of boot devices IB lt ver gt for a ConnectX device or gPXE for an Infini Host III device The priority of this list can be modified through BIOS setup A 7 Operation A 7 1 Prerequisites Make sure that your client is connected to the server s The BoIB image is already programmed on the adapter card see Section A 2 Start the Subnet Manager as described in Section A 4 Configure and start the DHC
113. e of sdp zcopy thresh is 64KB but is may be too low for some systems You will need to experiment with your hardware to find the best value page 62 libsdp so Library libsdp so is a dynamically linked library which is used for transparent integration of applica tions with SDP The library is preloaded and therefore takes precedence over glibc for certain socket calls Thus it can transparently replace the TCP socket family with SDP socket calls The library also implements a user level socket switch Using a configuration file the system administrator can set up the policy that selects the type of socket to be used libsdp so also has the option to allow server sockets to listen on both SDP and TCP interfaces The various configura tions with SDP TCP sockets are explained inside the etc libsdp conf file Configuring SDP To load SDP upon boot edit the file etc infiniband openib conf and set SDP LOAD yes 53 Mellanox Technologies Rev 1 10 54 SDP 6 3 1 Note For the changes to take effect run etc init d openibd restart SDP shares the same IP addresses and interface names as IPoIB See IPoIB configuring in Section 3 3 and Section 3 3 3 How to Know SDP Is Working Since SDP is a transparent TCP replacement it can sometimes be difficult to know that it is work ing correctly The sdpnetstat program can be used to verify both that SDP is loaded and is being used hostl sdpnetstat S
114. e of SLES 10 SP2 OS This section provides an example of installing the SLES 10 SP2 operating system on an iSCSI tar get and booting from a diskless machine via BoIB Note that the procedure described below assumes the following Theclient s LAN card is recognized during installation The iSCSI target can be connected to the client via LAN and InfiniBand Prerequisites See Section A 7 1 on page 154 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 161 Warning The following procedure modifies critical files used in the boot procedure It must be executed by users with expertise in the boot process Improper applica tion of this procedure may prevent the diskless machine from booting Procedure Step 1 Load the SLES 10 SP2 installation disk and enter the following parameters as boot options netsetup 1 WithISCSI 1 Boot from Hard Disk Installation Installation ACPI Disabled Installation Local APIC Disabled Installation Safe Settings Rescue System Memory Test Boot Options netsetup i HithISCSI 1 Step 2 Continue with the procedure as instructed by the installation program until the iSCSI Ini tiator Overview window appears Preparation 2e iSCSI Initiator Overview Language License Agreement Disk Activation e System Analysis Connected Targets e Time Zone Portal Address Target Name Start Up Installation e Installation Summary Perform In
115. e type ConnectX FW Version 2 6 000 Device ID 25418 Chip Revision AO Description Node Porti Port2 Sys image GUIDs 0002c90300001038 0002c90300001039 0002c9030000103a 0002c9030000103b MACs 0002c9001039 0002c900103a Board ID n a MT 04A0110002 VSD n a PSID MT 04A0110002 Assuming that BoIB is connected via Port 1 then the Port GUID is 00 02 c9 03 00 00 10 39 Mellanox Technologies Rev 1 10 152 Extracting the Port GUID Method II An alternative method for obtaining the port GUID involves booting the client machine via BoIB This requires having a Subnet Manager running on one of the machines in the InfiniBand subnet The 8 bytes can be captured from the boot session as shown in the figure below Mellanox ConnectX B ver 3 v2 0 000 gPXE 0 9 6 Open Source Boot Firmware 51 02 00 0 open 0 TXE 0 0 0 k up 0 ok Placing Client Identifiers in etc dhcpd confn The following is an excerpt of a etc dhcpd conf example file showing the format of represent ing a client machine for the DHCP server host host1 next server 11 4 3 7 filename pxelinux 0 fixed address 11 4 3 130 option dhcp client identifier 00 02 9 03 00 00 10 39 j A 3 1 2 For InfiniHost III Family Devices PCI Device IDs 25204 25218 When a BoIB client boots it sends the DHCP server various information including its DHCP cli ent identifier This identifier is used to
116. ections with them daemon mode Discover reachable SRP Targets given an infiniband HCA name and port rather than just by dev umad lt N gt where lt N gt is a digit Enable High Availability operation together with Device Mapper Multipath Havea configuration file that determines the targets to connect to Mellanox Technologies Rev 1 10 68 7 2 4 1 srp daemon commands equivalent to ibsrpdm srp daemon a o is equivalent to ibsrpdm srp daemon c a o is equivalent to ibsrpdm c Note These srp daemon commands can behave differently than the equivalent ibsrpdm command when etc srp daemon conf is not empty Srp daemon extensions to ibsrpdm To discover SRP Targets reachable from the HCA device lt InfiniBand HCA name gt and the port port num and to generate output suitable for echo you may execute hostl srp daemon c a o i lt InfiniBand HCA name gt p port number Note To obtain the list of InfiniBand HCA device names you can either use the ibstat tool or run 15 sys class infiniband To both discover the SRP Targets and establish connections with them just add the e option to the above command Executing srp daemon over a port without the a option will only display the reachable targets via the port and to which the initiator is not connected If executing with the e option it is better to omit a tis recommended to use the n option This option adds t
117. ed link modes Not reported Advertised auto negotiation No Speed Unknown 10000 Duplex Full Port Twisted Pair PHYAD 0 Transceiver internal Auto negotiation off Supports Wake on d Wake on d Current message level 0x00000000 0 Link detected yes Bonding Driver EoIB uses the standard Linux bonding driver For more information on the Linux Bonding driver please refer to lt kernel source gt Documentation networking bonding txt Currently not all bonding modes are supported LACP is not supported Jumbo Frames EoIB supports jumbo frames up to the InfiniBand limit of 4Kbytes To configure EoIB to work with jumbo frames you need to configure the entire InfiniBand fabric to use 4K MTU This includes configuring the SM InfiniBand switches and configuring the ConnectX HCA To configure the HCA port to work with 4K MTU set the mlx4 core module parameter set 4k mtu For how to configure the SM and switches refer to their corresponding documentation Module Parameters The mlx4 vnic driver supports the following module parameters These parameters are intended to enable more specific configuration of the mlx4 vnic driver to customer needs The mlx4 vnic is also effected by module parameters of other modules like the set 4k mtu of mlx4 core but these will not be addressed here The available module parameters include tx rings num Number of TX rings used per vNic use 0 for cores default 0 rx rings num Number of RX rin
118. ell in the face of faults It has been shown that for both regular and irregular topologies LASH outperforms Up Down The reason for this is that LASH distributes the traffic more evenly through a network avoiding the bottleneck issues related to a root node and always routes shortest path The algorithm was developed by Simula Research Laboratory Use R lash Q option to activate the LASH algorithm Note QoS support has to be turned on in order that SL VL mappings are used Note LMC gt 0 15 not supported by the LASH routing If this is specified the default routing algorithm is invoked instead DOR Routing Algorithm The Dimension Order Routing algorithm is based on the Min Hop algorithm and so uses shortest paths Instead of spreading traffic out across different paths with the same shortest distance it chooses among the available shortest paths based on an ordering of dimensions Each port must be consistently cabled to represent a hypercube dimension or a mesh dimension Paths are grown from a destination back to a source using the lowest dimension port of available paths at each step This provides the ordering necessary to avoid deadlock When there are multiple links between any two switches they still represent only one dimension and traffic is balanced across them unless port equalization is turned off In the case of hypercubes the same port must be used throughout the fabric to represent the hypercube dimension and match on
119. er qp log number of RDMARC buffers per QP default is 4 max is 7 log num cq log maximum number of CQs per HCA default is 16 max is 19 log num mcg log maximum number of multicast groups per HCA default is 13 max is 21 log num mpt log maximum number of memory protection table entries per HCA default is 17 max is 20 log num mtt log maximum number of memory translation table seg ments per HCA default is 20 max is 20 log num mac log maximum number of MACs per ETH port 1 7 int log num vlan log maximum number of VLANs per ETH port 0 7 int Mellanox Technologies Rev 1 10 208 use prio Enable steering by VLAN priority on ETH ports 0 1 default 0 bool 2 mlx4 ib Parameters debug level Enable debug tracing if gt 0 default 0 mlx4 en Parameters rss Use XOR hash function for RSS 0 default is xor rss mask RSS hash type bitmask default is Oxf num lro Number of LRO sessions per ring or disabled 0 default is 32 pfctx Priority based Flow Control policy on TX 7 0 Per priority bit mask default is 0 pfcrx Priority based Flow Control policy on RX 7 0 Per priority bit mask default is 0 inline thold Threshold for using inline data default is 128 4 mlx4 fc Parameters log exch per vhba Max outstanding FC exchanges per virtual HBA log Default 9 int max vhba per port Max vHBAs allowed per port Default 2 int Rev 1 10 Mellanox Te
120. f you do not wish to stop and restart the driver m1x4 en can be loaded by running sbin modprobe mlx4 en The result is a new net device appearing in ifconfig a Unloading the Driver If etc infiniband openib conf had MLX4 EN LOAD yes at driver start up then you can unload the m1x4 en driver by running etc init d openibd stop Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual Otherwise unload m1x4 en by running gt modprobe r mlx4 en 2 3 4 Ethernet Driver Usage and Configuration Toassign an IP address to the interface run gt ifconfig eth lt n gt ip where x is the OS assigned interface number To check driver and device information run gt ethtool i eth lt n gt Example gt ethtool i eth2 driver mlx4 en MT 04A0140005 version 1 4 0 March 2009 firmware version 2 6 000 bus info 0000 13 00 0 To query stateless offload status run gt ethtool k eth lt n gt To set stateless offload status run gt ethtool K eth lt n gt rx onloff tx on off sg onloff tso onjoff To query interrupt coalescing settings run gt ethtool c eth lt n gt By default the driver uses adaptive interrupt moderation for the receive path which adjusts the moderation time according to the traffic pattern To enable disable adaptive interrupt modera tion use the following command gt ethtool C eth lt n gt adaptive rx onloff Above
121. fabric links ibdiagnet pkey A dump of the the existing partitions and their member host ports ibdiagnet mcg A dump of the multicast groups their properties and member host ports ibdiagnet db A dump of the internal subnet database This file can be loaded in later runs using the load db option In addition to generating the files above the discovery phase also checks for duplicate node port GUIDs in the IB fabric If such an error is detected it is displayed on the standard output After the discovery phase is completed directed route packets are sent multiple times according to the c option to detect possible problematic paths on which packets may be lost Such paths are explored and a report of the suspected bad links is displayed on the standard output After scanning the fabric if the r option is provided a full report of the fabric qualities is dis played This report includes SM report Number of nodes and systems Hop count information maximal hop count an example path and a hop count histogram All CA to CA paths traced Credit loop report mgid mlid HCAs multicast group and report Partitions report IPoIB report Note In case the IB fabric includes only one CA then CA to CA paths are not reported Fur thermore if a topology file is provided ibdiagnet uses the names defined in it for the output reports 11 3 5 ERROR CODES 1 Failed to fully discover the fabric 2 Failed to parse c
122. formation for the specified device May specify more than one device lt port gt Optional but ports of the spec Print information for the specified port only of the specified requires specifying a ified device device device name Examples 1 List the status of all available InfiniBand devices and their ports gt ibstatus Infiniband device mlx4 0 port 1 status default gid fe80 0000 0000 0000 0000 0000 0007 3896 base lid 0x3 sm lid 0x3 state 4 ACTIVE phys state 5 LinkUp rate 20 Gb sec 4X DDR Infiniband device mlx4 0 port 2 status default gid fe80 0000 0000 0000 0000 0000 0007 3897 base lid 0 1 sm lid 0 1 state 4 ACTIVE phys state 5 LinkUp rate 20 Gb sec 4X DDR Infiniband device mthca0 port 1 status default gid fe80 0000 0000 0000 0002 c900 0101 d151 base lid 0x0 sm lid 0x0 state 2 INIT phys state 5 LinkUp rate 10 Gb sec 4X Infiniband device mthca0 port 2 status default gid fe80 0000 0000 0000 0002 c900 0101 d152 base lid 0x0 sm lid 0x0 state 2 INIT phys state 5 LinkUp rate 10 Gb sec 4X Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 126 2 List the status of specific ports of specific devices gt ibstatus mthca0 1 mlx4 0 2 Infiniband device mthca0 port 1 status default gid fe80 0000 0000 0000 0002 c900 0101 d151 base lid 0x0 sm lid 0 0 state 2 INIT phys state
123. g simple sysfs opera tions Alternatively it can be handled automatically by the dcbxd daemon if the interface is con nected to an FCoE switch supporting DCBX negotiation of the FCoE feature e g Cisco Nexus Once a vHBA is instantiated on an Ethernet interface it immediately attempts to log into the FC fabric Provided that the FC fabric and FC targets are well configured LUNs will map to SCSI disk devices dev sdX XX vHBAs instantiated automatically by the dcbxd daemon are created on a VLAN 0 interface with VLAN priority set to the value negotiated with the switch This takes advantage of PFC which allows pausing FCoE traffic when needed without pausing the entire Ethernet link Also with proper configuration of the FCoE switch the link s maximum band width can be divided as needed between FCoE and regular Ethernet traffic Instantiating VHBAs manually allows creating them on VLAN interfaces with any arbitrary VLAN id and priority as well as on the regular without VLAN Ethernet interfaces Using the regular interface means that PFC cannot be used In this case it is highly recommended that both the FCoE switch and the m1x4 en driver be con figured to use link pause regular flow control Otherwise any FCoE packet drop will trigger SCSI errors and timeouts FCoE Configuration After installation please edit the file etc mlxfc mlxfc conf and set the following variables FC SPEC set to T11 or pre T11 as supporte
124. ger human provides a set of rules policy that define how the network is being configured and how its resources are split to different QoS Levels The policy also define how to decide which QoS Level each application or ULP or service use 2 The SM analyzes the provided policy to see if it is realizable and performs the necessary fabric setup Part of this policy defines the default QoS Level of each partition The SA is enhanced to match the requested Source Destination QoS Class Service ID PKey against the policy so cli ents ULPs programs can obtain a policy enforced QoS The SM may also set up partitions with appropriate IPoIB broadcast group This broadcast group carries its QoS attributes SL MTU RATE and Packet Lifetime IPoIB is being setup IPoIB uses the SL MTU RATE and Packet Lifetime available on the mul ticast group which forms the broadcast group of this partition 4 MPI which provides non IB based connection management should be configured to run using hard coded SLs It uses these SLs for every being opened ULPs that use CM interface like SRP have their own pre assigned Service ID and use it while obtaining PathRecord MultiPathRecord PR MPR for establishing connections The SA receiv ing the PR MPR matches it against the policy and returns the appropriate PR MPR including SL MTU RATE and Lifetime ULPs and programs e g SDP use CMA to establish RC connection provide the CMA the
125. ges and var log opensm log The first file will register only general major events whereas the second file will include details of reported errors All errors reported in this second file should be treated as indicators of IB fabric health issues Note that when a fatal and non recoverable error occurs opensm will exit Both log files should include the message SUBNET UP if opensm was able to setup the subnet correctly 10 2 1 Syntax opensm OPTIONS where OPTIONS are version Prints OpenSM version and exits F config config file The name of the OpenSM config file If not specified etc opensm opensm conf will be used if it exists Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 88 c create config file name gt g l Pr guid lt GUID OpenSM will dump its configuration to the specified file and exit This is one way to generate an OpenSM configuration file template in hexadecimal gt This option specifies the local port GUID value with which OpenSM should bind OpenSM may be bound to 1 port at a time If the GUID given is 0 OpenSM displays a list of possible port GUIDs and waits for user input Without g OpenSM tries to use the default port lmc LMC value This option specifies th subnet s LMC value The number of LIDs assigned to each port is 2 LMC The LMC value must be in th rang 0 7 LMC values gt 0 allow multiple paths between ports L
126. ght be mapped to Each set holds SL and optionally Max MTU Max Rate Packet Lifetime and Path Bits Note Path Bits are not implemented in BXOFED IV Matching Rules A list of rules that match an incoming PR MPR request to a QoS Level The rules are processed in order such as the first match is applied Each rule is built out of a set of match expressions which should all match for the rule to apply The matching expressions are defined for the following fields SRC and DST to lists of port groups Service ID to a list of Service ID values or ranges QoS Class to a list of QoS Class values or ranges CMA features The CMA interface supports Service ID through the notion of port space as a prefix to the port number which is part of the sockaddr provided to rdma resolve add The CMA also allows the ULP like SDP to propagate a request for a specific QoS Class The CMA uses the provided QoS Class and Service ID in the sent PR MPR IPoIB IPoIB queries the SA for its broadcast group information and uses the SL MTU RATE and Packet Lifetime available on the multicast group which forms this broadcast group SDP SDP uses CMA for building its connections The Service ID for SDP is 0x000000000001 PPPP where PPPP are 4 hexadecimal digits holding the remote TCP IP Port Number to connect to RDS RDS uses CMA and thus it is very close to SDP The Service ID for RDS is 0x000000000106PPPP where PPPP are 4 hexadecimal digits holding
127. gs use 0 for cores default 0 The receive rings service all vNics that use the HCA port e mcast create Create multicast group during join request default 0 int If set mlx4 vnic will request the mcast group to be created and not only Joined The creation operation is usually the responsibility of the BridgeX eport state enfroce Bring vNIC link indication up only when corresponding External Port 1s up default 0 Mellanox Technologies Rev 1 10 EoIB 50 lro num Number of LRO sessions per ring or disable 0 default 32 napi weight NAPI weight default 32 max tx outs Max outstanding TX packets default 16 linear small pkt Use linear buffer for small packets default 1 If set causes packet copy for small packets net admin Network administration enabled default 1 If disabled no network administered interfaces will be opened Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 51 5 RDS 5 1 Overview Reliable Datagram Sockets RDS is a socket API that provides reliable in order datagram delivery between sockets over RC or TCP IP RDS is intended for use with Oracle RAC 11g For programming details enter host1 man rds 52 RDS Configuration The RDS ULP is installed as part of Mellanox BXOFED for Linux To load the RDS module upon boot edit the file etc infiniband openib conf and set RDS LOAD yes Note For the changes to take effect run
128. h Description Commands h Print the help menu hh Print an extended help menu d evice device All Specify the device to which the Flash is connected guid lt GUID gt burn sg GUID base value 4 GUIDs are automatically assigned to the following values guid gt node GUID guid 1 gt portl guid 2 gt port2 guid 3 gt system image GUID Note Port2 guid will be assigned even for a single port HCA the HCA ignores this value guids lt GUIDs gt burn sg 4 GUIDs must be specified here The specified GUIDs are assigned the following values repectively node portl port2 and system image GUID Note Port2 guid must be specified even for a single port HCA the HCA ignores this value It can be set to 0 0 mac lt MAC gt burn sg MAC address base value Two MACs are automatically assigned to the following values gt portl 1 gt port2 Note This switch is applicable only for Mellanox Technologies Ethernet products macs lt MACs gt burn sg Two MACs must be specified here The specified MACs are assigned to port and port2 repectively Note This switch is applicable only for Mellanox Technologies Ethernet products blank_guids burn Burn the image with blank GUIDs and MACs where applicable These values can be set later using the sg command see Table 34 below clear_semaphore No commands Force clear the Flash semaphore on the device No command is allowed when this switch allowed i
129. he form of vconfig Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 45 4 3 5 1 Configuring VLANs 4 3 6 4 3 7 4 3 8 To configure VLAN tag for a vNic just add the VLAN tag property to the configuration file in host administrated mode or configure the vNic on the appropriate vHub in network administered mode In host administered mode if a vHub with the requested VLAN tag is not available yet it will most likely be created automatically Host administered VLAN configuration in centralized conf file Add vid lt vlan tag gt or remove vid property for no VLAN Host administered VLAN configuration with ifcfg ethX configuration files VNICVLAN lt vlan tag gt or remove VNICVLAN property for no VLAN Note Using a VLAN tag value of 0 is not recommended because the traffic using it would not be separated form non VLAN traffic EoIB Multicast Configuration Configuring Multicast for EoIB interfaces is identical to multicast configuration for native Ethernet interfaces Note EoIB maps Ethernet multicast addresses to InfiniBand MGIDs Multicast GID The map is done in a way that ensures that different vHubs use mutually exclusive MGIDs This prevents vNics on different vHubs from communicating with one another EoIB and QoS EoIB enables the use of InfiniBand service levels At this time EoIB supports the use of a single SL for all vNics The configuration of the SL is performed t
130. he initiator ext to the connecting string See Section 7 2 5 for more details srp daemon has a configuration file that can be set where the default is etc srp daemon conf Use the f to supply a different configuration file that configures the targets srp daemon is allowed to connect to The configuration file can also be used to set values for additional parameters e g max cmd per lun max sect Acontinuous background daemon operation providing an automatic ongoing detection and con nection capability See Section 7 2 4 Automatic Discovery and Connection to Targets Make sure that the ib srp module is loaded the SRP Initiator can reach an SRP Target and that an SM is running To connect to all the existing Targets in the fabric run srp daemon e o This utility will scan the fabric once connect to every Target 1t detects and then exit Note srp daemon will follow the configuration it finds in etc srp daemon conf Thus it will ignore a target that is disallowed in the configuration file To connect to all the existing Targets in the fabric and to connect to new targets that will join the fabric execute srp daemon e This utility continues to execute until it is either killed by the user or encounters connection errors such as no SM in the fabric To execute SRP daemon as a daemon you may run run srp daemon found under usr sbin providing it with the same options used for running srp daemon Note Make
131. hine file Includes the list of machines Example host1 cat home lt username gt cluster host1 Mellanox Technologies Rev 1 10 MPI 76 host2 host1 8 5 2 Bandwidth Test Performance To run the OSU Bandwidth test enter hostl usr mpi gcc mvapich mvapich ver bin mpirun rsh np 2 hostfile home lt username gt cluster usr mpi gcc mvapich mvapich ver tests osu benchmarks osu ver osu bw OSU MPI Bandwidth Test v3 0 Size Bandwidth MB s 1 4 62 2 8 91 4 17 70 8 32 59 16 60 13 32 113 21 64 194 22 128 293 20 256 549 43 512 883 23 1024 1096 65 2048 1165 60 4096 1233 91 8192 1230 90 16384 1308 92 32768 1414 75 65536 1465 28 131072 1500 36 262144 1515 26 524288 1525 20 1048576 1527 63 2097152 1530 48 4194304 1537 50 8 5 3 Latency Test Performance To run the OSU Latency test enter hostl usr mpi gcc mvapich lt mvapich ver gt bin mpirun rsh np 2 hostfile home lt username gt cluster usr mpi gcc mvapich mvapich ver tests osu benchmarks osu ver osu latency OSU MPI Latency Test v3 0 H Size Latency us 0 1 20 1 1 21 2 1 21 4 1 21 8 1 23 16 1 24 32 1 33 64 1 49 128 2 66 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 256 3 08 512 3 61 1024 4 82 2048 6 09 4096 8 62 8192 13 59 16384 18 12 32768 28 81 65536 50 38 131072 93 70 262144 178 77 524288 349 31 1048576 689 25 2097152 1371 04 4194304 273
132. hnologies Rev 1 10 MPI 74 Generating public private rsa key pair Enter file in which to save the key home username ssh id rsa Enter passphrase empty for no passphrase Enter same passphrase again Your identification has been saved in home lt username gt ssh id_rsa Your public key has been saved in home username ssh id rsa pub The key fingerprint is 38 1b 29 d 4 08 00 4a 0e 50 0 05 44 e7 9 05 lt username gt hostl Step 2 Check that the public and private keys have been generated hostl cd home lt username gt ssh host1 1s host1 1s la total 40 drwx 2 root root 4096 Mar 5 04 57 drwxr x 13 root root 4096 Mar 4 18 27 1 root root 1675 Mar 5 04 57 rsa rw r r 1 root root 404 Mar 5 04 57 id rsa pub Step 3 Check the public key host1 cat id rsa pub ssh rsa AAAAB3NzaClyc2EAAAABIWAAAQEAl1zVY8VBHOh90kZN7OAlibUQ74RXm4zHeczyVxpYHaDPyDmq ezbYMKrCIVzdlObH ZkCOrpLYviUOOoUHd3fvNTfMs0gcGg08PysUf l2FyYjira2Pl1xyg6mkHLG GqVutfEMmABZ3wNCUg6J2X3G uiuSWXeubZmbXcMrP wAIWByfH8ajwo6A5WioNbFZElbYeeNfPZf4UNCgMOAMWp64sL58tkt32F RGmyLXOWZL27Synsn 6dHpxMqBorXNCOZBe4kTnUqm63nQ2zliqVMdL9FrCmalxlIOu94 SQJAjwONevaMzFKEHe7YHg6YrN fXunfdbEurzB524TpPcrodZlfCQ username hostl Step 4 Now you need to add the public key to the authorized keys2 file on the target machine host1 cat id rsa pub xargs ssh host2 echo gt gt home lt username gt ssh authorized_ keys
133. hrough the BridgeX Refer to BridgeX doc umentation for the use of non default SL IP Configuration Based on DHCP Setting an EoIB interface configuration based on DHCP v3 1 2 which is available via www isc org is performed similarly to the configuration of Ethernet interfaces In other words you need to make sure that EoIB configuration files include the following line For RedHat BOOTPROTO dhcp For SLES BOOTPROTO dchp Note If EoIB configuration files are included ifcfg eth lt n gt files will be installed under etc sys config network scripts on a RedHat machine etc sysconfig network on a SuSE machine Mellanox Technologies Rev 1 10 EoIB 46 4 3 8 1 DHCP Server No special configuration is needed to use a DHCP server with EoIB The DHCP server can run on a server which is located on the Ethernet side using any Ethernet HW or on a server located on the InfiniBand side and running EoIB module 4 3 9 Static EoIB Configuration If you wish you can use an EoIB configuration that is not based on DHCP Static configuration is performed in a similar fashion to a typical Ethernet device See your Linux distribution documenta tion for additional information about configuring IP addresses Note Ethernet configuration files are located at etc sysconfig network scripts on a RedHat machine etc sysconfig network on a SuSE machine 4 4 Sub Interfaces VLAN EoIB interfaces do not support creating sub in
134. i fying vf 0 disables all messages Specifying vf OxFF enables all messages see V High verbosity levels may require increasing the transaction timeout with the t option Display this usage info then exit To run osmtest in the default mode simply enter 1 osmtest The default mode runs all the flows except for the Quality of Service flow see Section 10 6 After installing opensm and if the InfiniBand fabric is stable it is recommended to run the fol lowing command in order to generate the inventory file 1 osmtest f c Immediately afterwards run the following command to test opensm 1 osmtest f a Finally it is recommended to occasionally run osmtest v with verbosity to verify that noth ing in the fabric has changed Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 97 10 4 Partitions OpenSM enables the configuration of partitions PKeys in an InfiniBand fabric By default OpenSM searches for the partitions configuration file under the name usr etc opensm par titions conf To change this filename you can use opensm with the Pconfig or P flags The default partition is created by OpenSM unconditionally even when a partition configuration file does not exist or cannot be accessed The default partition has a P Key value of Ox7fff The port out of which runs OpenSM is assigned full membership in the default partition
135. iSER application with a specific target TCP IP port range IPoIB with a default PKey IPoIB with a specific Any ULP application with a specific Service ID in the PR MPR query Any ULP application with a specific PKey in the PR MPR query Any ULP application with a specific target IB port GUID in the PR MPR query Since any section of the policy file is optional as long as basic rules of the file are kept such as no referring to nonexisting port group having default QoS Level etc the simple policy section qos ulps can serve as a complete QoS policy file The shortest policy file in this case would be as follows qos ulps default 0 Zdefault SL end qos ulps It is equivalent to the previous example of the shortest policy file and it is also equivalent to not having policy file at all Below is an example of simple QoS policy with all the possible keywords qos ulps default 0 default SL sdp port num 30000 0 SL for application running on top of SDP when a destination TCP IPport is 30000 sdp port num 10000 20000 0 sdp 1 default SL for any other application running on top of SDP rds 2 4 SL for RDS traffic Iser port num 900 0 SL for iSER with a specific target port iser 3 default SL for iSER ipoib pkey 0x0001 0 SL for IPoIB on partition with pkey 0x0001 ipoib 4 default IPoIB partition pkey Ox7FFF any service id 0x6234 6 match PR MPR query
136. ice Mellanox Boot over IB implements a network driver with IP over IB acting as the transport layer IP over IB is part of the Mellanox BXOFED for Linux software package see www mellanox com The binary code is exported by the device as an expansion ROM image Supported Mellanox HCA Devices and Firmware Table 35 Supported Mellanox Technologies Devices and PCI Device IDs Device Name hs Firmware MT25408 ConnectX IB SDR PCI Express 2 0 2 5GT s 25408 0 6340 fw 25408 MT25408 ConnectX IB DDR PCI Express 2 0 2 5GT s 25418 0x634a fw 25408 MT25408 ConnectX IB DDR PCI Express 2 0 5 0GT s 26418 0x6732 fw 25408 MT25408 ConnectX IB QDR PCI Express 2 0 5 0GT s 26428 0x673c fw 25408 MT25208 InfiniHost III Ex 25218 0x6282 fw 25218 MT25204 InfiniHost III Lx 25204 0x6274 fw 25204 Tested Platforms See the Boot over IB Release Notes boot over ib release notes txt BoIB in Mellanox BXOFED The Boot over IB binary files are provided as part of BXOFED The following binary files are included 1 PXE binary files for Mellanox HCA devices HCA Single Dual port ConnectX IB SDR PCI DevID 25408 Mellanox Technologies Rev 1 10 150 AJ CONNECTX IB 25408 ROM X X XXX rom HCA Single Dual port ConnectX IB DDR PCI DevID 25418 CONNECTX IB 25418 ROM X X XXX rom HCA Single Dual port ConnectX IB DDR amp PCI Express 2 0 5 0GT s PCI DevID
137. id pci cr0 conf0 2 The firmware mlx file w ID X X XXX mlx 3 One of the expansion ROM binary files listed in Section B 1 3 Firmware burning example The following command burns a firmware image and an expansion ROM image to the Flash device of a ConnectX adapter card mlxburn dev dev mst mt25448 pci cr0 fw fw 25408 X X XXX mlx conf MNEH28 XTC ini exp rom ConnectX EN 25448 ROM X X XXX rom Preparing the DHCP Server in Linux Environment The DHCP server plays a major role in the boot process by assigning IP addresses for ConnectX EN PXE clients and instructs the clients where to boot from When the ConnectX EN PXE boot session starts the PXE firmware attempts to bring up a Con nectX network link port If it succeeds to bring up a connected link the PXE firmware communi cates with the DHCP server The DHCP server assigns an IP address to the PXE client and provides it with the location of the boot program Configuring the DHCP Server B 3 1 1 For ConnectX Family Devices When a ConnectX EN PXE client boots it sends the DHCP server various information including its DHCP hardware Ethernet address MAC The MAC address is 6 bytes long and it is used to distinguish between the various DHCP sessions Extracting the MAC Address Method I Run the following commands Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 177 Note The following MFT commands assume tha
138. ies Mellanox BXOFED Stack for Linux User s Manual Revision History Rev 1 10 February 23 2010 First release Mellanox Technologies Rev 1 10 11 12 Preface This Preface provides general information concerning the scope and organization of this User s Manual It includes the following sections Intended Audience page 12 Documentation Conventions page 12 Related Documentation page 14 Intended Audience This manual is intended for system administrators responsible for the installation configuration management and maintenance of the software and hardware of VPI InfiniBand Ethernet FCoE FCoIB systems comprising servers with adapter cards VPI gateways and InfiniBand switch plat forms It is also intended for application developers Documentation Conventions Typographical Conventions Table 19 Typographical Conventions Description Convention Example File names file extension Directory names directory Commands and their parameters command parami Optional items i Mutually exclusive parameters f pl p2 l ps Optional mutually exclusive parameters pl p2 p3 Prompt of a user command under bash shell hostname Prompt of a root command under bash shell hostname Prompt of a user command under tcsh shell tcsh Environment variables VARIABLE Code example if a b Comment at the beginning of a code line LH Characters
139. ifies the time in milliseconds used for transaction timeouts Specifying t 0 disables timeouts Without t OpenSM defaults to a timeout value of 200 milli seconds This option defines the log to be the given file By default the log goes to var log osm log For the log to go to standard output use f stdout Mellanox Technologies Rev 1 10 96 OpenSM Subnet Manager v verbose V vf h help 10 3 2 Running osmtest This option increases the log verbosity level The v option may be specified multiple times to further increase the verbosity level S the vf option for more information about log verbosity This option sets the maximum verbosity level and forces log flushing The V is equivalent to vf OxFF d 2 See the vf option for more information about log verbosity This option sets the log verbosity level A flags field must follow the D option bit set clear in the flags enables disables a specific log level as follows BIT LEVEL ENABLED 0x01 ERROR error messages 0x02 INFO basic messages low volume 0x04 VERBOSE interesting stuff moderate volume 0x08 DEBUG diagnostic high volume 0x10 FUNCS function entry exit very high volume 0x20 FRAMES dumps all SMP and GMP frames 0x40 ROUTING dump FDB routing information 0x80 currently unused Without vf osmtest defaults to ERROR INFO 0x3 Spec
140. ig Show Release Help Abort Step 19 At the end of the file copying stage the Finishing Basic Installation window will pop up and ask for confirming a reboot You can click OK to skip count down See image below Note Assuming that the machine has been correctly configured to boot from ConnectX EN PXE via its connection to the ISCSI target make sure that MLNX EN has the highest priority in the BIOS boot sequence Mellanox Technologies Rev 1 10 194 Preparation v Language v License Agreement v System Analysis E Time Zone Copy files to installed system Installation v Installation Summary Save configuration gt Perform Installation Install boot manager Configuration Save installation settings e Hostname e Root Password Prepare system for initial boot e Network e Customer Center e Online Update e Service Users e Clean Up e Release Notes The system will reboot now e Hardware Configuration Finished Dm Step 20 Once the boot is complete the Startup Options window will pop up Select SUSE Linux Enterprise Server 10 then press Enter SUSE Linux Enterprise Server 10 Floppu SUSE Linux Enterprise Server 10 Failsafe Boot Options Step 21 The Hostname and Domain Name window will pop up Continue configuring your machine until the operating system is up then you can start running the machine in normal operation mode Rev 1 10 Mellanox Technologies Me
141. il the algorithm works as follows 1 LASH determines the shortest path between all pairs of source destination switches Note LASH ensures the same SL is used for all SRC DST DST SRC pairs and there is no guaran tee that the return path for a given DST SRC will be the reverse of the route SRC DST 2 LASH then begins an SL assignment process where a route is assigned to a layer SL if the addition of that route does not cause deadlock within that layer This is achieved by main taining and analysing a channel dependency graph for each layer Once the potential addition of a path could lead to deadlock LASH opens a new layer and continues the process 1 List of compute nodes CNs be specified by u or cn guid file OpenSM options Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 10 5 6 10 5 7 3 Once this stage has been completed it is highly likely that the first layers processed will con tain more paths than the latter ones To better balance the use of layers LASH moves paths from one layer to another so that the number of paths in each layer averages out Note that the implementation of LASH in opensm attempts to use as few layers as possible This number can be less than the number of actual layers available In general LASH is a very flexible algorithm It can for example reduce to Dimension Order Routing in certain topologies it is topology agnostic and fares w
142. in Section A 3 1 Configuring the DHCP Server and connected to the client machine 3 An initrd file 4 To add an IB driver into initrd you need to copy the IB modules to the diskless image Your machine needs to be pre installed with BXOFED Adding the IB Driver to the initrd File Warning The following procedure modifies critical files used in the boot procedure It must be executed by users with expertise in the boot process Improper applica tion of this procedure may prevent the diskless machine from booting Step 1 Back up your current initrd file Step 2 Make a new working directory and change to it Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual Step 3 Step 4 Step 5 Step 6 Step 7 Step 8 157 host1 mkdir tmp initrd ib host1 cd tmp initrd ib Normally the initrd image is zipped Extract it using the following command host1 gzip dc lt initrd image cpio id The initrd files should now be found under tmp initrd ib Create a directory for the InfiniBand modules and copy them hostl mkdir p tmp initrd ib lib modules ib hostl cd lib modules uname r updates kernel drivers hostl cp infiniband core ib addr ko tmp initrd ib lib modules ib hostl cp infiniband core ib core ko tmp initrd ib lib modules ib hostl cp infiniband core ib mad ko tmp initrd ib lib modules ib hostl cp infiniband core ib sa ko tmp initrd ib lib modules ib hostl cp infiniban
143. in the re transmission of pack ets and therefore in bandwidth loss This check should be conducted for each port after the driver is loaded To check for symbol errors enter cat sys class infiniband lt device gt ports 1 counters symbol error Mellanox Technologies Rev 1 10 198 The command above is performed Port 1 of the device lt device gt The output value should be 0 if no symbol errors were recorded Bandwidth is expected to vary between systems It heavily depends on the chipset memory and CPU Nevertheless the full wire speed should be achieved by the host With IB SDR the expected unidirectional full wire speed bandwidth is 900MB sec With IB DDR and PCI Express Gen 1 the expected unidirectional full wire speed bandwidth is 1400MB sec See Section C 1 With IB DDR and PCI Express Gen 2 the expected unidirectional full wire speed bandwidth is 1800MB sec See Section C 1 With IB QDR and PCI Express Gen 2 the expected unidirectional full wire speed bandwidth is 3000MB sec See Section C 1 To check the adapter s maximum bandwidth use the ib write bw utility To check the adapter s latency use the ib write lat utility Note The utilities ib write ib write lat are installed as part of Mellanox BXOFED Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 199 Appendix D ULP Performance Tuning D 1
144. in this default mode opensm will scan the IB fabric initialize it and sweep occasionally for changes To run opensm in the default mode simply enter hostis opensm Note that opensm needs to be run on at least one machine in an IB subnet By default an opensm run is logged to two files var log messages and var log opensm log The first file message registers only general major events the second file opensm log includes details of reported errors All errors reported in opensm log should be treated as indicators of IB fabric health Both log files should include the message SUBNET UP if opensm was able to setup the subnet correctly Note If a fatal non recoverable error occurs opensm exits 10 2 4 1 Running OpenSM As Daemon OpenSM can also run as daemon To run OpenSM in this mode enter host1 etc init d opensmd start Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 94 10 3 osmtest Description osmtest is a test program for validating the InfiniBand Subnet Manager and Subnet Administrator osmtest provides a test suite for opensm It can create an inventory file of all available nodes ports and PathRecords including all their fields It can also verify the existing inventory with all the object fields and matches it to a pre saved one See Section 10 3 2 osmtest has the following test flows Multicast Compliancy test Event Forwarding test Service Record registr
145. inition 11 2 3 Addressing 11 3 ibdiagnet IB Net Diagnostic 11 3 1 SYNOPSYS 11 3 2 Output Files 11 3 3 ERROR CODES 11 4 ibdiagpath IB diagnostic path 11 4 1 SYNOPSYS 11 4 2 Output Files 11 4 3 ERROR CODES 11 5 devices 11 6 ibv devinfo 11 7 ibstatus 11 8 ibportstate 11 9 ibroute 11 10 smpquery 11 11 perfquery 11 12 ibcheckerrs 11 13 mstflint 11 14 ibv asyncwatch Appendix A Boot over IB BoIB A Overview A 2 Burning the Expansion ROM Image A 3 Preparing the DHCP Server in Linux Environment A 4 Subnet Manager OpenSM A 5 TFTP Server A 6 BIOS Configuration A 7 Operation A 8 Diskless Machines A 9 iSCSI Boot 102 103 103 104 105 105 105 106 107 107 109 111 111 111 111 112 112 112 113 114 114 115 116 117 117 117 117 118 118 118 118 120 120 121 121 122 122 122 123 124 126 131 135 138 141 143 147 149 149 150 150 154 154 154 154 156 159 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual A 10 WinPE 173 Appendix B ConnectX EN PXE 175 B 1 Overview 175 B 2 Burning the Expansion ROM Image 176 B 3 Preparing the DHCP Server in Linux Environment 176 B 4 TFTP Server 178 B 5 BIOS Configuration 178 B 6 Operation 178 B 7 Diskless Machines 179 B 8 iSCSI Boot 180 2 9 iSCSI Boot Example of SLES 10 SP2 OS 182 B 10 WinPE 195 Appendix C Performance Troubleshooting 197 C 1 PCI Express Performance Troubleshooting 197
146. int debug messages Examples To print SDP usage per connect and listern to STDERR include the following statement log min level 7 destination stderr A non root user can configure 1ibsdp so to record function calls and return values in the file tmp libsdp log lt pid gt root log goes to var log libsdp log for this example by including the following statement in Libsdp conf log min level 2 destination file libsdp log To print errors only to syslog include the following statement log min level 9 destination syslog To print maximum output to the file tmp sdp_debug log lt pid gt include the following state ment log min level 1 destination file sdp_debug log Mellanox Technologies Rev 1 10 56 Kernel Space SDP Debug The SDP kernel module can log detailed trace information if you enable it using the debug level variable in the sysfs filesystem The following command performs this hostl echo 1 gt sys module ib sdp debug level Note Depending on the operating system distribution on your machine you may need an extra level parameters in the directory structure so you may need to direct the echo command to sys module ib sdp parameters debug level Turning off kernel debug is done by setting the sysfs variable to zero using the following command hostl echo 0 gt sys module ib sdp debug level To display debug information use the dmesg command hostl dmesg 6 4 Environment Variables
147. ion System Analysis e Time Zone Installation e Installation Summary e Perform Installation Configuration e Root Password Hostname Network Customer Center Online Update Service Users Clean Up Release Notes Hardware Configuration Step 9 Select the appropriate Region and Time Zone in the Clock and Time Zone window then click Finish Installation Select Mode New Installation v Language v License Agreement v Disk Activation v System Analysis gt Time Zone Installation e Installation Summary e Perform Installation Configuration Hostname Root Password Network Customer Center Online Update Service Users Clean Up Release Notes Hardware Configuration Egg Clock and Time Zone Region Time Zone Canada Central and South America Russia Asia Australia Africa Pacific Global Etc Alaska Aleutian Arizona Central East Indiana Hawaii Indiana Starke Michigan Mountain Pacific Samoa Hardware Clock Set To Abort 2 Date 07 52 06 24 03 2008 Finish Mellanox Technologies Rev 1 10 165 166 Step 10 In the Installation Settings window click Partitioning to get the Suggested Partitioning window Preparation v Language v License Agreement v Disk Activation v v Installation Settings System Analysis Click any head
148. ion e Installation Summary e Perform Installation IP Address Port Configuration e Root Password Hostname Network X No Authentication Customer Center Online Update Service Users Clean Up Username Password LJ Incoming Authentication Release Notes Hardware Configuration J C Outgoing Authentication Username Password Help Back Abort Mellanox Technologies Rev 1 10 184 Step 4 Details of the discovered iSCSI target s will be displayed in the iSCSI Initiator Discovery window Select the target that you wish to connect to and click Connect A iSCSI Initiator Discovery License Agreement gt Disk Activation e System Analysis Portal Address Target Name Connected e Time Zone 10 4 3 7 3260 1 iqn 2007 08 7 3 4 10 iscsiboot False Installation e Installation Summary Perform Installation Configuration Root Password Hostname e Network e Customer Center e Online Update Service e Users e Clean Up e Release Notes e Hardware Configuration Connect If no iSCSI target was recognized then either the target was not properly installed or no connection was found between the client and the iSCSI target Open a shell to ping the iSCSI target you can use CTRL ALT F2 and verify that the target is or is not accessible To return to the graphical installation screen press CTRL ALT F7 E Rev 1 10 Mellanox Technologies
149. ion script with a configuration file using the n option containing the full IP configuration The IPoIB configuration file can specify either or both of the following data for an IPoIB interface static IPoIB configuration An IPoIB configuration based on an Ethernet configuration Note See your Linux distribution documentation for additional information about configuring IP addresses The following code lines are an excerpt from a sample IPoIB configuration file Static settings all values provided by this file IPADDR ib0 11 4 3 175 NETMASK ib0 255 255 0 0 NETWORK ib0 11 4 0 0 BROADCAST ib0 11 4 255 255 ONBOOT ib0 1 Based on eth0 each will be replaced with a corresponding octet from eth0 LAN INTERFACE ib0 eth0 IPADDR ib0 11 4 NETMASK ib0 255 255 0 0 NETWORK ib0 11 4 0 0 BROADCAST ib0 11 4 255 255 ONBOOT ib0 1 Based on the first eth lt n gt interface that is found for n 0 1 each will be replaced with a corresponding octet from eth lt n gt Mellanox Technologies Rev 1 10 IPoIB 32 LAN INTERFACE ib0 IPADDR 1b0 11 4 NETMASK 160 255 255 0 0 NETWORK 1b0 11 4 0 0 BROADCAST 1b0 11 4 255 255 ONBOOT ib0 1 3 3 3 3 4 3 4 1 Manually Configuring IPoIB To manually configure IPoIB for the default IB partition VLAN perform the following steps Note This manual configuration persists only until the next reboot or driver restart Step 1 To configure
150. ional verbosity vvv or v v v V ersion Optional Show version info D irect Optional Use directed path address arguments The path is a comma sepa rated list of out ports Examples 0 self port 0 1 2 1 4 out via port 1 then 2 G uid Optional Use GUID address argument In most cases it is the Port GUID Example 0x08f1040023 s lt smlid gt Optional Use lt smlid gt as the target lid for SM SA queries C ca name Optional Use the specified channel adapter or router P ca port Optional Use the specified port t timeout ms Optional Override the default timeout for the solicited MADs msec dest dr path lid Optional Destination s directed path LID or GUID guid lt portnum gt Optional Destination s port number lt op gt lt value gt Optional query Define the allowed port operations enable disable reset speed and query In case of multiple channel adapters CAs or multiple ports without a CA port being specified a port is chosen by the utility according to the following criteria 1 The first ACTIVE port that is found 2 If not found the first port that is UP physical link state is LinkUp Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 128 Examples 1 Query the status of Port 1 of CA mlx4 0 using ibstatus and use its output LID 3 in this case to obtain additional link information using ibportsta
151. is configured as Eth it may also function as a Fibre Channel HBA Port Type Management ConnectX ports can be individually configured to work as InfiniBand or Ethernet or Fibre Channel over Ethernet ports By default both ConnectX ports are initialized as InfiniBand ports If you wish to change the port type use the connectx port config script after the driver is loaded Running sbin connectx port config s will show current port configuration for all Con nectX devices Port configuration is saved in the file etc infiniband connectx conf This saved configu ration is restored at driver restart only if restarting via etc init d openibd restart Possible port types are eth Ethernet 16 Infiniband Table 22 lists the ConnectX port configurations supported by VPI Table 22 Supported ConnectX Port Configurations Port 1 Configuration Port 2 Configuration ib ib ib eth eth eth Note that the configuration Port eth and Port2 ib is not supported Also note that FCoE can run only on a port configured as eth and the m1x4 en driver must be loaded The port link type can be configured for each device in the system at run time using the sbin connectx port config script This utility will prompt for the PCI device to be modified 1f there is only one it will be selected automatically In the next stage the user will be prompted for the desired mode for each port The desired po
152. istrationSupported 0 0000 LinkSpeedA ctive 5 0 Gbps LinkSpeedEnabled 2 5 Gbps or 5 0 Gbps FilterRawOutb Mkey Violations xe PkeyViolations 0 QkeyViolations 0 GuidCap 128 ClientReregister SubnetTimeout 18 RespTimeVal 16 LocalPhysErr OverrunErr Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 138 2 Query SwitchInfo by GUID gt smpquery G switchinfo 0x000b8cffff004016 Switch info Lid 3 InboundPartEnf 1 OutboundPartEnf 1 3 Query Nodelnfo by direct route gt smpquery D nodeinfo 0 Node info DR path slid 65535 dlid 65535 0 Base Verse ee 1 1 55 1 Node ee Channel Adapter NumpPorts 2 SystemGuid 0x0002c9030000103b uid 0x0002c90300001038 ROK Guide nee 0x0002c90300001039 A 128 0x634a RGVISION 0x000000a0 IEocalBOnt5eee a 1 Wend onde eee tre 0x0002c9 11 11 perfquery Applicable Hardware All InfiniBand devices Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 139 Description Queries InfiniBand ports performance and error counters Optionally it displays aggregated coun ters for all ports of a node It can also reset counters after reading them
153. ity arbitration tables and further it can be listed in both tables The limit of high priority VLArb table qos type high limit indicates the number of high priority packets that can be transmitted without an opportunity to send a low priority packet Specifically the number of bytes that can be sentis high limit times 4K bytes A high limit value of 255 indicates that the byte limit is unbounded Note If the 255 value is used the low priority VLs may be starved A value of 0 indicates that only a single packet from the high priority table may be sent before an opportunity 1s given to the low priority table Keep in mind that ports usually transmit packets of size equal to MTU For instance for 4KB MTU a single packet will require 64 credits so in order to achieve effective VL arbitration for packets of 4KB MTU the weighting values for each VL should be multiples of 64 Below is an example of SL2VL and VL Arbitration configuration on subnet qos ca max vls 15 qos ca high limit 6 qos ca vlarb high 0 4 qos ca vlarb low 0 0 1 64 2 128 3 192 4 0 5 64 6 64 7 64 qos ca sl2v1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 7 qos swe max vls 15 qos swe high limit 6 qos swe vlarb high 0 4 qos swe vlarb low 0 0 1 64 2 128 3 192 4 0 5 64 6 64 7 64 qos swe sl2vl 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 7 In this example there are 8 VLs configured on subnet VLO to VL7 VLO is defined as a high prior ity VL and it is limited to 6 x 4KB 24KB in a single
154. ke the burn process burn two images on Flash The current default failsafe burn pro cess burns a single image in alternating locations v Print version info Table 34 mstflint Commands Command Description b urn Burn Flash q uery Query miscellaneous Flash firmware characteristics v erify Verify the entire Flash bb Burn Block Burn the given image as is without running any checks sg Set GUIDs ri lt out file gt Read the firmware image on the Flash into the specified file dc lt out file gt Dump Configuration Print a firmware configuration file for the given image to the spec ified output file e rase lt addr gt Erase sector rw lt addr gt Read one DWORD from Flash ww lt addr gt lt data Write one DWORD to Flash wwne lt addr gt Write one DWORD to Flash without sector erase wbne lt addr gt size data Write a data block to Flash without sector erase rb lt addr gt size out file Read a data block from Flash swreset SW reset the target InfniScale IV device This command is supported only in the In Band access method Possible command return values are 0 successful completion 1 error has occurred 7 the burn command was aborted because firmware is current Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 146 Examples 1 Find Mellanox Technologies s ConnectX amp VPI cards with PCI Express running at 2 5GT s and InfiniBand ports at DDR or Ethernet ports at 10GigE gt
155. le host1 opensm R file M opensm lid matrix dump The dump file is named opensm lid matrix dump and will be generated in the standard opensm dump directory var log by default when OM LOG ROUTING logging flag is set When routing engine file is activated but the dump file is not specified or cannot be opened the default lid matrix algorithm will be used There is also a switch forwarding tables dumper which generates a file compatible with dump lfts sh output This file can be used as input for forwarding tables loading by file routing engine Both or one of options U and M can be specified together with R file Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 10 6 10 6 1 10 6 2 105 Quality of Service Management in OpenSM Overview When Quality of Service QoS in OpenSM is enabled using the Q or qos flags OpenSM looks for a QoS Policy file During fabric initialization and at every heavy sweep OpenSM parses the QoS policy file applies its settings to the discovered fabric elements and enforces the provided policy on client requests The overall flow for such requests is as follows The request is matched against the defined matching rules such that the QoS Level definition is found Given the QoS Level a path s search is performed with the given restrictions imposed by that level Figure 5 QoS Manager Administrator InfiniBa
156. line to make changes or use the Change menu below Time Zone Overview Expert Installation gt Installation Summary Keyboard Layout e Perform Installation English US Configuration e Hostname Partitioning eop pasaar Create boot partition dev sda1 70 5 MB with ext2 e Network Create swap partition dev sda2 502 0 MB Customer Center Create root partition dev sda3 7 4 GB with reiserfs e Online Update e Service Software e Users e Clean Up SUSE Linux Enterprise Server 10 e Release Notes ctc WIndawsystem 3 GNOME Desktop Environment for Server Hardware Configuration 4 Server Base System Novell AppArmor 4 Print Server Size of Packagesto Install 1 3 GB Language Primary Language English US Show Release Notes Help Abort Step 11 Select Base Partition Setup on This Proposal then click Next Your hard disks have Suggested Partitioning been checked The partition setup displayed is proposed for your hard drive Create boot partition dev sdal 70 5 MB with ext2 Create swap partition dev sda2 502 0 MB To accept these Create root partition dev sda3 7 4 GB with reiserfs suggestions and continue select Accept Proposal Ifthe suggestion does not fit your needs create your own partition setup starting with the partitions as currently present on the disks For this select Custom Partition Setup This is also the option to choose
157. llanox BXOFED Stack for Linux User s Manual 195 Step 22 Optional If you wish to have the second instance of connecting to the iSCSI Target go through the Ethernet driver copy the initrd file under boot to a new location add the Ethernet driver into it after the load commands of the iSCSI Initiator modules and con tinue as described in Section B 7 on page 179 Warning Pay extra care when changing initrd as any mistake may prevent the client machine from booting It is recommended to have a back up iSCSI Initiator on a machine other than the client you are working with to allow for debug in case initrd gets corrupted Next edit the init file that is in the initrd zip and look for the following string if ISCSI TARGET IPADDR then iscsiserver iSCSI TARGET IPADDR fi Now add before the string the following line iSCSI TARGET IPADDR Ethernet IP Address of iSCSI Target Example iSCSI TARGET _IPADDR 11 4 3 7 Also edit the file boot grub menu 1st and delete the following string ibft mode off 10 WinPE Mellanox ConnectX EN PXE enables WinPE boot via TFTP For instructions on preparing a WinPE image please see http etherboot org wiki winpe Mellanox Technologies Rev 1 10 196 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 197 Appendix C Performance Troubleshooting C A C 2 PCI Express Performance Troubleshooting For the bes
158. logy file name gt 2 Define the environment variable IBDIAG FILE Mellanox Technologies Rev 1 10 117 InfiniBand Fabric Diagnostic Utilities 118 11 2 2 11 2 3 11 3 To specify the local system name to an diagnostic tool use one of the following two options 1 On the command line specify the system name using the option s 1ocal system name gt 2 Define the environment variable IBDIAG SYS NAME IB Interface Definition The diagnostic tools installed on a machine connect to the IB fabric by means of an HCA port through which they send MADs To specify this port to an IB diagnostic tool use one ofthe follow ing options 1 On the command line specify the port number using the option p local port number gt see below 2 Define the environment variable IBDIAG PORT NUM In case more than one HCA device is installed on the local machine it is necessary to specify the device s index to the tool as well For this use on of the following options 1 On the command line specify the index of the local device using the following option 1 index of local device gt 2 Define the environment variable IBDIAG DEV IDX Addressing Note This section applies to the ibdiagpath tool only A tool command may require defining the destination device or port to which it applies The following addressing modes can be used to define the IB ports Using a Directed Route to the destination
159. mImageGuid then the previous LID of the same LMC group If none prefer those which go through another NodeGuid Fall back to the number of paths method 1f all go to same node 10 5 1 Effect of Topology Changes OpenSM will preserve existing routing in any case where there is no change in the fabric switches unless the r reassign lids option is specified r reassign lids This option causes OpenSM to reassign LIDs to all end nodes Specifying r on a running subnet may disrupt subnet traf fic Without r OpenSM attempts to preserve existing LID assignments resolving multiple use of same LID Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 100 If a link is added or removed OpenSM does not recalculate the routes that do not have to change A route has to change if the port is no longer UP or no longer the MinHop When routing changes are performed the same algorithm for balancing the routes is invoked In the case of using the file based routing any topology changes are currently ignored The file routing engine just loads the LFTs from the file specified with no reaction to real topology Obvi ously this will not be able to recheck LIDs by GUID for disconnected nodes and LFTs for non existent switches will be skipped Multicast is not affected by file routing engine this uses min hop tables 10 5 2 Min Hop Algorithm The Min Hop algorithm is invoked when neither UPDN or the file method are s
160. me 0 049 ms 64 bytes from 11 4 3 176 icmp seq 4 ttl 64 time 0 065 ms 11 4 3 176 ping statistics 5 packets transmitted 5 received 0 packet loss time 3999ms rtt min avg max mdev 0 044 0 058 0 079 0 014 ms pipe 2 The ib bonding Driver The ib bonding driver is a High Availability solution for IPoIB interfaces It is based on the Linux Ethernet Bonding Driver and was adapted to work with IPoIB The ib bonding package contains a bonding driver and a utility called ib bond to manage and control the driver operation The ib bonding driver comes with the ib bonding package run rpm qi ib bonding to get the package information Using the ib bonding Driver The ib bonding driver can be loaded manually or automatically Manual Operation Use the utility ib bond to start query or stop the driver For details on this utility please read the documentation for the ib bonding package under usr share doc ib bonding 0 9 0 ib bonding txt on RedHat and usr share doc packages ib bonding 0 9 0 ib bonding txt on SuSE Automatic Operation There are two ways to configure automatic ib bonding operation 1 Using the openibd configuration file as described in the following steps Edit the file etc infiniband openib conf to define bonding parameters Example Enable the bonding driver on startup IPOIBBOND ENABLE yes Set bond interface names IPOIB_BONDS bond0 bond8007 Set specific bond params address and slaves
161. mentation on how to configure a network administered vNic To disable network administered vnics on the host side load mlx4 vnic module with the net admin module parameter set to 0 VLAN Configuration As explained in the topology section a vNic instance is associated with a specific vHub group This vHub group is connected to a BridgeX external port and has a VLAN tag attribute When creating configuring a vNic you define the VLAN tag it will use via the vid or the VNICVLAN fields If these fields are absent the vnic will not have a VLAN tag If the vNic has a VLAN tag it will be present in all EoIB packets sent by the vNics and will be verified on all packets received on the vNic When passed from the InfiniBand side to the Ethernet side the EoIB encapsulation will be stripped but the VLAN tag will remain For example if the vNic eth23 is associated with a vHub that uses BridgeX bridge01 eport A10 and VLAN tag 8 all incoming and outgoing traffic on eth23 will use a VLAN tag of 8 This will be enforced by both BridgeX and destination hosts When a packet is passed from the internal fabric to the Ethernet subnet through the BridgeX it will have a true Ethernet VLAN tag of 8 The VLAN implementation used by EoIB uses OS un aware VLANs This is in many ways similar to switch tagging in which an external Ethernet switch adds strips tags on traffic preventing the need of OS intervention EoIB does not support OS aware in t
162. modules mlx4 core low level helper mlx4 ib IB mlx4 en Ethernet mlx4 fc FCoE and mlx4 vnic EoIB Mid layer core Verbs MADs SA CM CMA uVerbs uMADs Upper Layer Protocols ULPs IPoIB RDS SDP SRP Initiator MPI Open MPI stack supporting the InfiniBand interface OSU MVAPICH stack supporting the InfiniBand interface MPI benchmark tests OSU BW LAT Intel MPI Benchmark Presta OpenSM InfiniBand Subnet Manager Utilities Diagnostic tools Performance tests Documentation 1 4 Architecture Figure 1 shows a diagram of the Mellanox BXOFED stack and how upper layer protocols ULPs interface with the hardware and with the kernel and user spaces The application level also shows the versatility of markets that BXOFED applies to Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual Figure 1 BXOFED Stack Diagnostics OpenSM Sockets Layer umad API UDP ICMP IPoIB mlx4 en pxu ib_core HUA 4 ib HCA Driver mlx4 core Markets ES Linux Linux Modules in OFED Currently Not Supported by Mellanox Applications HE OFED The following sub sections briefly describe the various components of the BXOFED stack 1 4 1 mthea HCA IB Driver mthea is the low level driver implementation for the following Mellanox Technologies HCA InfiniBand devices InfiniHost InfiniHost III Ex and InfiniHost III Lx 1 4 2 mlx4 VPI Driver mlx4 is the low le
163. mp new init ib img At this stage the modified initrd including the IB driver is ready and located at tmp new init ib img gz Copy itto the original initrd location and rename it properly iSCSI Boot Mellanox Boot over IB enables an iSCSI boot of an OS located on a remote iSCSI Target It has a built in iSCSI Initiator which can connect to the remote iSCSI Target and load from it the kernel and initrd There are two instances of connection to the remote iSCSI Target the first is for get ting the kernel and initrd via BoIB and the second is for loading other parts of the OS via ini trd Note Linux distributions such as SuSE Linux Enterprise Server 10 SPx and Red Hat Enterprise Linux 5 1 or above can be directly installed on an iSCSI target At the end of this direct instal lation initrd is capable to continue loading other parts of the OS on the iSCSI target Other dis tributions may also be suitable for direct installation on iSCSI targets If you choose to continue loading the OS after boot through the HCA device driver please verify that the initrd image includes the HCA driver as described in Section A 8 Configuring an iSCSI Target in Linux Environment Prerequisites Step 1 Make sure that an iSCSI Target is installed on your server side Tip You can download and install an iSCSI Target from the following location http sourceforge net project showfiles php group 1d 108475 amp package id 117141 Step 2 Dedicate
164. n in var log srp daemon log 7 2 7 Shutting Down SRP SRP can be shutdown by using rmmod ib or by stopping the BXOFED driver etc init d openibd stop or as a by product of a complete system shutdown Prior to shutting down SRP remove all references to it The actions you need to take depend on the way SRP was loaded There are three cases 1 Without High Availability When working without High Availability you should unmount the SRP partitions that were mounted prior to shutting down SRP 2 After Manual Activation of High Availability If you manually activated SRP High Availability perform the following steps a Unmount all SRP partitions that were mounted b Killthe SRP daemon instances c Make sure there are no multipath instances running If there are multiple instances wait for them to end or kill them d Run multipath F 5 After Automatic Activation of High Availability If SRP High Availability was automatically activated SRP shutdown must be part of the driver shutdown etc init d openibd stop which performs Steps 2 4 of case b above However you still have to unmount all SRP partitions that were mounted before driver shutdown Mellanox Technologies Rev 1 10 SRP 72 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 8 8 1 8 2 8 2 1 73 Overview Mellanox BXOFED for Linux includes the following MPI implementations
165. n timeout with the t option d debug This option specifies a debug option These options are not normally needed The number following d selects the debug option to enable as follows OPT Description 0 Ignore other SM nodes di Force single threaded dispatching d2 Force log flushing after each log message d3 Disable multicast support Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 10 2 2 10 2 3 10 2 4 h help Display this usage info then exit Display this usage info then exit Environment Variables The following environment variables control open sm behavior OSM TMP DIR Controls the directory in which the temporary files generated by opensm are created These files are opensm subnet lst opensm fdbs and opensm mcfdbs By default this directory is var log OSM CACHE DIR opensn stores certain data to the disk such that subsequent runs are consistent The default directory used is var cache opensm The following file is included in it guid21id stores the LID range assigned to each GUID Signaling When opensm receives a HUP signal it starts a new heavy sweep as if a trap has been received or a topology change has been found Also SIGUSRI can be used to trigger a reopen of var log opensm log for logrotate pur poses Running opensm The defaults of opensm were designed to meet the common case usage on clusters with up to a few hundred nodes Thus
166. nal side are called external ports or eports Every BridgeX that 1s in use with EoIB needs to have one or more eports connected External ports eports and GW The combination of a specific BridgeX box and a specific eport is referred to as a gateway GW The GW is an entity that is visible to the EoIB host driver and is used in the configuration of the network interfaces on the host side For example in host administered vnics the user will request to open an interface on a specific GW identifying it by the BridgeX box and eport name Distinguish ing between GWs is important because they determine the network topology and affect the path that a packet traverses between hosts A packet that is sent from the host on a specific EoIB inter face will be routed to the Ethernet subnet through a specific external port connection on the BridgeX box Virtual Hubs vHubs Virtual hubs connect zero or more EoIB interfaces on internal hosts and an eport through a virtual hub Each vHub has a unique virtual LAN VLAN id Virtual hub participants can send packets to one another directly without the assistance of the Ethernet subnet external side routing This means that two EoIB interfaces on the same vHub will communicate solely using the InfiniBand fabric EoIB interfaces residing on two different vHubs whether on the same GW or not can not communicate directly There are two types of vHub a default vHub one per GW which has no VLAN ID and vHubs
167. nd subnet with l HDL OFED 1 3 based nodes There two ways to define QoS policy Advanced the advanced policy file syntax provides the administrator various ways to match a PathRecord MultiPathRecord PR MPR request and to enforce various QoS constraints on the requested PR MPR Simple the simple policy file syntax enables the administrator to match PR MPR requests by various ULPs and applications running on top of these ULPs Advanced QoS Policy File The QoS policy file has the following sections I Port Groups denoted by port groups This section defines zero or more port groups that can be referred later by matching rules see below Port group lists ports by Port GUID Port name which is a combination of NodeDescription and IB port number PKey which means that all the ports in the subnet that belong to partition with a given PKey belong to this port group Partition name which means that all the ports in the subnet that belong to partition with a given name belong to this port group Node type where possible node types are CA SWITCH ROUTER ALL and SELF SM s port Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 106 II QoS Setup denoted by gos setup This section describes how to set up SL2VL and VL Arbitration tables on various nodes in the fab ric However this is not supported in BXOFED SL2VL and VLArb tables should be configured in
168. nd any root nodes The list of the root nodes found by this auto detect stage is used by the ranking process stage Note The user can override the node list manually Note If this stage cannot find any root nodes and the user did not specify a guid list file OpenSM defaults back to the Min Hop routing algorithm 2 Ranking process All root switch nodes found in stage 1 are assigned a rank of 0 Using the BFS algorithm the rest of the switch nodes in the subnet are ranked incrementally This ranking aids in the process of enforcing rules that ensure loop free paths Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 3 Min Hop Table setting after ranking is done a BFS algorithm is run from each CA or switch node in the subnet During the BFS process the FDB table of each switch node traversed by BFS is updated in reference to the starting node based on the ranking rules and guid values At the end of the process the updated FDB tables ensure loop free paths through the subnet Note Up Down routing does not allow LID routing communication between switches that are located inside spine switch systems The reason is that there is no way to allow a LID route between them that does not break the Up Down rule One ramification of this is that you can not run SM on switches other than the leaf switches of the fabric 10 5 3 1UPDN Algorithm Usage 10 5 4 Activation through OpenSM Us
169. nd can control each flow utilization of fabric resources The InfiniBand Architecture Specification defines several hardware features and management interfaces for supporting QoS Up to 15 Virtual Lanes VL carry traffic in a non blocking manner Arbitration between traffic of different VLs is performed by a two priority level weighted round robin arbiter The arbiter is programmable with a sequence of VL weight pairs and a maximal number of high priority credits to be processed before low priority 1s served Packets carry class of service marking in the range 0 to 15 in their header SL field Each switch can map the incoming packet by its SL to a particular output VL based on a pro grammable table VL SL to VL MA P in port out port SL The Subnet Administrator controls the parameters of each communication flow by providing them as a response to Path Record PR or MultiPathRecord MPR queries 83 Mellanox Technologies Rev 1 10 84 Quality of Service 9 2 9 3 DiffServ architecture IETF RFC 2474 amp 2475 is widely used in highly dynamic fabrics The fol lowing subsections provide the functional definition of the various software elements that enable a DiffServ like architecture over the Mellanox BXOFED software stack QoS Architecture QoS functionality is split between the SM SA CMA and the various ULPs We take the chronol ogy approach to describe how the overall system works 1 The network mana
170. nk down to existing VHBAs This is done using sysfs operations When using the pre T11 stack the sysfs directory is located at sys class mlx4 fc When using the T11 stack the sysfs directory is located at sys module fcoe Both directories contain the same entries In the following the sysfs directory will be referred to as SFCSYSFS To create a new vHBA on an Ethernet interface e g eth3 run gt echo eth3 gt FCSYSFS create To destroy a previously created on an interface e g eth3 run gt echo eth3 gt SFCSYSFS destroy To signal link up to an existing VHBA e g on eth3 run gt echo eth3 gt SFCSYSFS link up To signal link down to an existing vHBA e g on eth3 run gt echo eth3 gt SFCSYSFS link down 2 4 4 2 Creating VHBAs That Use To create a vHBA that uses the PFC feature it is required to configure the Ethernet driver to sup port PFC create a VLAN Ethernet interface assign it a priority and start a VHBA on the interface The following steps demonstrate the creation of such a vHBA To configure the mlx4 en Ethernet driver to support PFC add the following line to the file etc modprobe conf and restart the network driver options mlx4 en pfctx Oxff pferx Oxff To create a VLAN with an ID e g 55 on interface e g eth3 run Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual gt vconfig add eth3 55 gt
171. nsmod lib modules ipv6 ko echo loading IB driver sbin insmod lib modules ib ib addr ko sbin insmod lib modules ib ib core ko sbin insmod lib modules ib ib mad ko sbin insmod lib modules ib ib sa ko sbin insmod lib modules ib ib cm ko sbin insmod lib modules ib ib uverbs ko sbin insmod lib modules ib ib ucm ko sbin insmod lib modules ib ib umad ko sbin insmod lib modules ib tw cm ko sbin insmod lib modules ib rdma cm ko sbin insmod lib modules ib rdma ucm ko sbin insmod lib modules ib mlx4 core ko sbin insmod lib modules ib mlx4 ib ko sbin insmod lib modules ib ib mthca ko sbin insmod lib modules ib ib ipoib ko Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 9 9 1 159 Note In case of interoperability issues between iSCSI and Large Receive Offload LRO change the last command above as follows to disable LRO sbin insmod lib modules ib ib ipoib ko lro 0 Step 10 Now you can assign an IP address to your IB device by adding a call to ifconfig or to the DHCP client in the init file after loading the modules If you wish to use the DHCP client then you need to add a call to the DHCP client in the init file after loading the IB modules For example sbin dhclient cf sbin dhclient conf ib1 Step 11 Save the init file Step 12 Close initrd hostl1 cd tmp initrd ib 1 find cpio H newc o gt tmp new initrd ib img host1 gzip t
172. od t lt topo file gt Specifies the topology file name s lt sys name gt Specifies the local system name Meaningful only if a topology file is specified Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 122 i dev index p lt port num gt o out dir lw 1x 4x 12x ls lt 2 5 5 10 gt pm pc P lt PM lt Trash gt gt h help V version vars 11 4 2 Output Files Specifies the index of the device of the port used to connect to the IB fabric in case of multiple devices on the local system Specifies the local device s port number used to connect to the IB fabric Specifies the directory where the output files will be placed default tmp Specifies th xpected link width Specifies th xpected link speed Dump all the fabric links pm Counters into ibdiagnet pm Reset all the fabric links pmCounters If any of the provided pm is greater then its provided value print it to screen Prints the Prints the Prints the Table 25 ibdiagpath Output Files help page information version of the tool tool s environment variables and their values Output File Description ibdiagpath log A dump of all the application reports generated according to the provided flags ibdiagnet pm A dump of the Performance Counters values of the fabric links 11 4 5 ERROR CODES 1 The path traced is un healthy 2 Failed to
173. of host and network administered vNics EoIB Host Administered vNic In host administered mode vNics are configured using static configuration files located on the host side These configuration files define the number of vNics and the vHub that each host adminis tered vNic will belong to i e the vNic s BridgeX box eport and VLAN id properties The mlx4 vnic confd service 1s used to read these configuration files and pass the relevant data to the mlx4 vnic module Two forms of configuration files are supported a central configuration file mlx4 vnic conf and vNic specific configuration files ifcfg ethX X both supply the same func tionality If both forms of configuration files exist the central configuration file has precedence and only this file will be used Mellanox Technologies Rev 1 10 EoIB 42 4 3 1 1 Central configuration file etc infiniband mlx4 vnic conf The mlx4 vnic conf file consists of lines each describing one vNic The following file format 1s used name eth44 mac 00 25 8B 27 14 78 ib port mlx4 0 1 vid 0 vnic id 5 bx 00 00 00 00 00 00 03 B2 eport A10 name eth45 mac 00 25 8B 27 15 78 ib port mlx4 0 1 vnic id 6 bx 00 00 00 00 00 00 03 B2 eport A10 name eth47 mac 00 25 8B 27 16 84 ib port mlx4 0 1 vid 2 vnic id 7 bx BX001 eport A10 name eth40 mac 00 25 8B 27 17 93 ib port mlx4 0 1 vnic id 8 bx BX001 eport A10 The fields used in the file have the following meaning name mac i
174. ommand line options 3 Failed to intract with IB fabric 4 Failed to use local device or local port 5 Failed to use Topology File 6 Failed to load requierd Package Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 11 4 11 4 1 121 ibdiagpath IB diagnostic path ibdiagpath traces a path between two end points and provides information regarding the nodes and ports traversed along the path It utilizes device specific health queries for the different devices along the path The way ibdiagpath operates depends on the addressing mode used on the command line If directed route addressing is used d flag the local node is the source node and the route to the des tination port is known apriori On the other hand if LID route or by name addressing is employed then the source and destination ports of a route are specified by their LIDs or by the names defined in the topology file In this case the actual path from the local port to the source port and from the source port to the destination port is defined by means of Subnet Management Linear Forwarding Table queries of the switch nodes along that path Therefore the path cannot be predicted as it may change ibdiagpath should not be supplied with contradicting local ports by the p and d flags see syn opsis descriptions below In other words when ibdiagpath is provided with the options p and d together the first port in the
175. on and holds the IBNL files required to load this topology To use these files you will need to set the environment variable named IBDM IBNL PATH to that directory The directory is located in tmp or in the output directory provided by the o flag load db lt file name gt gt Load subnet data from the given db file and skip h help V version vars subnet discovery stage Note Some of the checks require actual subnet dis covery and therefore would not run when load db is Specified These checks are Duplicated zero guids link state SMs status Prints the help page information Prints the version of the tool Prints the tool s environment variables and their values Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 120 11 3 2 Output Files Table 24 ibdiagnet Output Files Output File Description ibdiagnet log A dump of all the application reports generate according to the provided flags ibdiagnet Ist List of all the nodes ports and links in the fabric ibdiagnet fdbs A dump of the unicast forwarding tables of the fabric switches ibdiagnet mcfdbs A dump of the multicast forwarding tables of the fabric switches ibdiagnet masks In case of duplicate port node Guids these file include the map between masked Guid and real Guids ibdiagnet sm List of all the SM state and priority in the fabric ibdiagnet pm A dump of the pm Counters values of the
176. on the Min Hop algorithm but avoids port equalization except for redundant links between the same two switches This provides deadlock free routes for hypercubes when the fabric is cabled as a hypercube and for meshes when cabled as a mesh OpenSM also supports a file method which can load routes from a table see Modular Routing Engine below The basic routing algorithm is comprised of two stages 1 MinHop matrix calculation How many hops are required to get from each port to each LID The algorithm to fill these tables is dif ferent if you run standard min hop or Up Down For standard routing a relaxation algorithm is used to propagate min hop from every destination LID through neighbor switches For Up Down routing a BFS from every target is used The BFS tracks link direction up or down and avoid steps that will per form up after a down step was used 2 Once MinHop matrices exist each switch is visited and for each target LID a decision is made as to what port should be used to get to that LID This step is common to standard and Up Down routing Each port has a counter counting the number of target LIDs going through it When there are multiple alternative ports with same MinHop to a LID the one with less previously assigned ports is selected If LMC gt 0 more checks are added Within each group of LIDs assigned to same target port Use only ports which have same MinHop First prefer the ones that go to different syste
177. on the adapter card 2 The DHCP server is installed and configured as described in Section B 3 1 Configuring the DHCP Server and connected to the client machine 3 An initrd file 4 To add an Ethernet driver into initrd you need to copy the Ethernet modules to the diskless image Your machine needs to be pre installed with BXOFED that is appropriate for the kernel version the diskless image will run Adding the Ethernet Driver to the initrd File Warning The following procedure modifies critical files used in the boot procedure It must be executed by users with expertise in the boot process Improper applica tion of this procedure may prevent the diskless machine from booting Step 1 Back up your current initrd file Step 2 Make a new working directory and change to it host1 mkdir tmp initrd en Mellanox Technologies Rev 1 10 180 B 8 host1 cd tmp initrd en Step 3 Normally the initrd image is zipped Extract it using the following command host1 gzip dc lt initrd image cpio id The initrd files should now be found under tmp initrd_en Step 4 Create a directory for the ConnectX EN modules and copy them hostl mkdir p tmp initrd en lib modules mlnx en host1 cd lib modules uname r updates kernel drivers hostl cp net mlx4 mlx4 core ko tmp initrd en lib modules mlnx en hostl cp net mlx4 mlx4 en ko tmp initrd en lib modules mlnx en Step 5 To load the modules you need the insmod execut
178. or Linux User s Manual 2 M U S a u m X ucast cache This option enables unicast routing cache and prevents rout ing recalculation which is a heavy task in a large clus ter when there was no topology change detected during the heavy sweep or when the topology change does not require new routing calculation e g when one or more CAs RTRs leaf switches going down or one or more of these nodes coming back after being down A very common case that is handled by the unicast routing cache is host reboot which otherwise would cause two full routing recalculations one when the host goes down and the other when the host comes back online connect roots This option enforces a routing engine currently up down only to make connectivity between root switches and in this way to be fully IBA complaint In many cases this can vio late pure deadlock free algorithm so use it carefully lid matrix file file name This option specifies the name of the lid matrix dump file from where switch lid matrices min hops tables will be loaded lfts file file name This option specifies th nam of the LFTs file from where switch forwarding tables will be loaded sadb file lt file name gt This option specifies the name of th SA DB dump file from where SA database will be loaded root guid file file name Set the root nodes for the Up Down or Fat T
179. or Linux User s Manual 10 OpenSM Subnet Manager 10 1 Overview OpenSM is an InfiniBand compliant Subnet Manager SM It is provided as a fixed flow execut able called opensm accompanied by a testing application called osmtest OpenSM implements an InfiniBand compliant SM according to the InfiniBand Architecture Specification chapters Man agement Model 13 Subnet Management 14 and Subnet Administration 15 10 2 opensm Description opensm is an InfiniBand compliant Subnet Manager and Subnet Administrator that runs on top of the Mellanox BXOFED stack opensm performs the InfiniBand specification s required tasks for initializing InfiniBand hardware One SM must be running for each InfiniBand subnet opensm also provides an experimental version of a performance manager opensm defaults were designed to meet the common case usage on clusters with up to a few hun dred nodes Thus in this default mode opensm will scan the IB fabric initialize it and sweep occasionally for changes opensm attaches to a specific IB port on the local machine and configures only the fabric con nected to it If the local machine has other IB ports opensm will ignore the fabrics connected to those other ports If no port is specified opensm will select the first best available port opensm can also present the available ports and prompt for a port number to attach to By default the opensm run is logged to two files var log messa
180. ork applications depend For more details see Chapter 6 SDP Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 1 4 6 1 4 7 1 4 8 1 4 9 1 5 SRP SRP SCSI RDMA Protocol is designed to take full advantage of the protocol offload and ROMA features provided by the InfiniBand architecture SRP allows a large body of SCSI software to be readily used on InfiniBand architecture The SRP driver known as the SRP Initiator differs from traditional low level SCSI drivers in Linux The SRP Initiator does not control a local HBA instead it controls a connection to an IO controller known as the SRP Target to provide access to remote storage devices across an InfiniBand fabric The SRP Target resides in an IO unit and provides storage services See Chapter 7 SRP MPI Message Passing Interface MPI is a library specification that enables the development of parallel software libraries to utilize parallel computers clusters and heterogeneous networks Mellanox BXOFED includes the following MPI implementations over InfiniBand Open MPI an open source MPI 2 implementation by the Open MPI Project OSU MVAPICH an MPI 1 implementation by Ohio State University Mellanox BXOFED also includes MPI benchmark tests such as OSU BW LAT Intel MPI Bench mark and Presta InfiniBand Subnet Manager All InfiniBand compliant ULPs require a proper operation of a Subnet Manager SM running
181. ot of the driver 1 Execute modprobe dm multipath 2 Execute modprobe ib srp 3 Make sure you have created file etc udev rules d 91 srp rules as described above 4 Execute for each port and each HCA srp daemon c e R 300 i lt InfiniBand HCA name gt p port number This step can be performed by executing srp daemon sh which sends its log to var log srp daemon log Now it is possible to access the SRP LUNs on dev mapper Note It is possible for regular non SRP LUNs to also be present the SRP LUNs may be iden tified by their names You can configure the etc multipath conf file to change multipath behavior Note It is also possible that the SRP LUNs will not appear under dev mapper This can occur if the SRP LUNs are in the black list of multipath Edit the blacklist section in etc multipath conf and make sure the SRP LUNs are not black listed Automatic Activation of High Availability Set the value of SRPHA ENABLE in etc infiniband openib conf to yes Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 71 Note For the changes in openib conf to take effect run etc init d openibd restart From the next loading of the driver it will be possible to access the SRP LUNS on dev mapper Note It is possible that regular not SRP LUNs may also be present the SRP LUNs may be identified by their name tis possible to see the output of the SRP daemo
182. ote iSCSI Target the first is for Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual B 8 1 181 getting the kernel and initrd via ConnectX EN and the second is for loading other parts of the OS via initrd Note Linux distributions such as SuSE Linux Enterprise Server 10 SPx and Red Hat Enterprise Linux 5 1 can be directly installed on an iSCSI target At the end of this direct installation initrd is capable to continue loading other parts of the OS on the iSCSI target Other distributions may also be suitable for direct installation on iSCSI targets If you choose to continue loading the OS after boot through the HCA device driver please verify that the initrd image includes the adapter driver as described in Section B 7 1 Configuring an iSCSI Target in Linux Environment Prerequisites Step 1 Make sure that an iSCSI Target is installed on your server side Tip You can download and install an iSCSI Target from the following location http sourceforge net project showfiles php group 1d 108475 amp package id 117141 Step 2 Dedicate a partition on your iSCSI Target on which you will later install the operating sys tem Step 3 Configure your iSCSI Target to work with the partition you dedicated If for example you choose partition dev sda5 then edit the iSCSI Target configuration file etc ietd conf to include the following line under the iSCSI Target iqn line Lun
183. over InfiniBand Open MPI an open source MPI 2 implementation by the Open MPI Project OSU MVAPICH an MPI 1 implementation by Ohio State University These MPI implementations along with MPI benchmark tests such as OSU BW LAT Intel MPI Benchmark and Presta are installed on your machine as part of the Mellanox BXOFED for Linux installation Table 23 lists some useful MPI links Table 23 Useful MPI Links MPI Standard http www unix mcs anl gov mpi Open MPI http www open mpl org MVAPICH MPI http nowlab cse ohio state edu projects mpi iba MPI Forum http www mpi forum org This chapter includes the following sections Prerequisites for Running MPI page 73 MPI Selector Which MPI Runs page 74 Compiling MPI Applications page 75 OSU MVAPICH Performance page 75 Open MPI Performance page 78 Prerequisites for Running MPI For launching multiple MPI processes on multiple remote machines the MPI standard provides a launcher program that requires automatic login 1 password less onto the remote machines SSH Secure Shell is both a computer program and a network protocol that can be used for logging and running commands on remote computers and or servers SSH Configuration The following steps describe how to configure password less access over SSH Step 1 Generate an ssh key on the initiator machine host1 host1 ssh keygen t rsa Mellanox Tec
184. p 7 The iSCSI Initiator Overview window will pop up Click Toggle Start Up to change start up from manual to automatic Click Finish Preparation Language License Agreement gt Disk Activation e System Analysis Service Connected Targets e Time Zone Portal Address s Target Name Start Up di Installation 10 4 3 7 60 1 iqn 2007 3 4 10 manual e Installation Summary e Perform Installation iSCSI Initiator Overview Configuration e Root Password e Hostname e Network e Customer Center e Online Update e Service e Users e Clean Up e Release Notes e Hardware Configuration Add Log Out Toggle Start Up Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual Step 8 Select New Installation then click Finish in the Installation Mode window Preparation v Language License Agreement Disk Activation System Analysis e Time Zone Installation e Installation Summary e Perform Installation Configuration e Root Password Hostname Network Customer Center Online Update Service Users Clean Up Release Notes Hardware Configuration Step 9 Select the appropriate Region and Time Zone in the Clock and Time Zone window then click Finish Installation Select Mode New Installation v Language v License Agreement v Disk Activation v Sys
185. parse command line options 3 More then 64 hops are required for traversing the local port to the Source port and then to the Destination port 4 Unable to traverse the LFT data from source to destination 5 Failed to use Topology File 6 Failed to load required Package 11 5 ibv devices Applicable Hardware All InfiniBand devices Description Lists InfiniBand devices available for use from userspace including node GUIDs Synopsis ibv devices Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 11 6 Examples 1 List the names of all available InfiniBand devices gt ibv devices device ibv devinfo Applicable Hardware All InfiniBand devices Description node GUID 0002c9000101d150 0000000000073895 Queries InfiniBand devices and prints about them information that 1s available for use from user space Synopsis ibv devinfo d lt device gt i lt port gt Table 26 lists the various flags of the command Table 26 ibv_devinfo Flags and Options E Fla Optional Default Description 8 Mandatory If Not Specified P d lt device gt Optional First found device _ Run the command for the provided IB device device ib dev lt device gt i port Optional All device ports Query the specified device port lt port gt ib port lt port gt 1 Optional Inactive Only list the names of InfiniBand devices list V O
186. pecified The Min Hop algorithm is divided into two stages computation of minhop tables on every switch and LFT output port assignment Link subscription is also equalized with the ability to override based on port GUID The latter is supplied by i lt equalize ignore guids file gt ignore guids lt equalize ignore guids file gt This option provides the means to define a set of ports by guid that will be ignored by the link load equalization algorithm Note that only endports CA switch port 0 and router ports and not switch external ports are supported LMC awareness routes based on remote system or switch basis 10 5 3 Purpose of UPDN Algorithm The UPDN algorithm is designed to prevent deadlocks from occurring in loops of the subnet A loop deadlock is a situation in which it is no longer possible to send data between any two hosts connected through the loop As such the UPDN routing algorithm should be used if the subnet is not a pure Fat Tree and one of its loops may experience a deadlock due for example to high pressure The UPDN algorithm is based on the following main stages 1 Auto detect root nodes based on the CA hop length from any switch in the subnet a statistical histogram is built for each switch hop num vs number of occurrences If the histogram reflects a specific column higher than others for a certain node then it is marked as a root node Since the algorithm is statistical it may not fi
187. pecifies the local port GUID value with which OpenSM should bind OpenSM may be bound to 1 port at a time If GUID given is 0 OpenSM displays a list of possible port GUIDs and waits for user input Without g OpenSM tries to use the default port This option displays a menu of possible local port GUID values with which osmtest could bind This option specifies the name of the inventory file Nor mally osmtest expects to find an inventory file which osmtest uses to validate real time information received from the SA during testing If i is not specified osmtest defaults to the file osmtest dat See c option for related information This option runs the specified stress test instead of the normal test suite Stress test options are as follows OPT Description sl Single MAD response SA queries s2 Multi MAD RMPP response SA queries s3 Multi MAD RMPP Path Record SA queries Without s stress testing is not performed Multicast ModeThis option specify length of Multicast test timeout log file OPT Description M1 Short Multicast Flow default single mode M2 Short Multicast Flow multiple mode M3 Long Multicast Flow single mode M4 Long Multicast Flow multiple mode Single mode Osmtest is tested alone with no other apps that interact with OpenSM MC Multiple mode Could be run with other apps using MC with OpenSM Without M default flow testing is performed This option spec
188. pertise in the boot process Improper applica tion of this procedure may prevent the diskless machine from booting Procedure Step 1 Load the SLES 10 SP2 installation disk and enter the following parameters as boot options netsetup 1 WithISCSI 1 Boot from Hard Disk Installation Installation ACPI Disabled Installation Local APIC Disabled Installation Safe Settings Rescue System Memory Test Boot Options netsetup i WithISCSI 1 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 183 Step 2 Continue with the procedure as instructed by the installation program until the iSCSI Ini tiator Overview window appears Preparation V Language License Agreement gt Disk Activation e System Analysis Service Connected Targets e Time Zone iSCSI Initiator Overview Portal Address Target Name Start Up Installation e Installation Summary Perform Installation Configuration Root Password Hostname Network Customer Center Online Update Service Users Clean Up Release Notes Hardware Configuration Step 3 Click the Add tab in the iSCSI Initiator Overview window An iSCSI Initiator Discovery window will pop up Enter the IP Address of your iSCSI target and click Next Preparation Language License Agreement gt Disk Activation e System Analysis e Time Zone 3 iSCSI Initiator Discovery Installat
189. poses a 20 Bytes HW address to the OS The mlx4 vnic module is designed for Mellanox s ConnectX family of HCAs and intended to be used with Mellanox s BridgeX gateway family Having a BridgeX gateway is a requirement for using EoIB It performs two operations Enables the layer 2 address translation required by the mlx4 vnic module Enables routing of packets from the InfiniBand fabric to a 1 or 10 GigE Ethernet subnet EoIB Topology EoIB is designed to work over an InfiniBand fabric In order for it to work properly it requires the presence of two entities Subnet Manager SM BridgeX gateway The required subnet manager configuration is similar to that of other InfiniBand applications and ULPs and is not unique to EoIB Refer to Chapter 10 OpenSM Subnet Manager for more information on SM and OpenSM Other then the subnet manager and the BridgeX you will most 39 Mellanox Technologies Rev 1 10 EoIB 40 4 2 1 4 2 2 4 2 3 likely need one or more InfiniBand switches and probably some Ethernet switches as well sim ple EoIB setup will look something like this The BridgeX gateway is at the heart of EoIB On one side usually referred to as the internal side it is connected to the InfiniBand fabric by one or more links On the other side usually referred to as the external side it is connected to the Ethernet subnet by one or more ports The Ethernet connections on the BridgeX s exter
190. proc scsi tgt trace level echo add mgmt dbg gt proc scsi tgt trace level Mellanox Technologies Rev 1 10 206 n echo add out of mem gt proc scsi_tgt trace_level E 3 How to Unload Shutdown 1 Unload ib srpt modprobe r ib srpt 2 Unload scst and its dev handlers first modprobe r scst_vdisk scst 3 Unload ofed etc rc d openibd stop Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 207 Appendix F mlx4 Module Parameters In order to set m1x4 parameters add the following line s to etc modpobe conf options mlx4 core parameter lt value gt and or options mlx4 ib parameter lt value gt and or options mlx4 parameter lt value gt and or options mlx4 fc parameter lt value gt The following sections list the available m1x4 parameters 1 mlx4 core Parameters set 4k mtu Attempt to set 4K MTU to all ConnectX ports int msi x Attempt to use MSI X if nonzero default 1 enable qos Enable Quality of Service support in the HCA if gt 0 default 0 block loopback Block multicast loopback packets if gt 0 default 1 internal err resetReset device on internal errors if non zero default 1 debug level Enable debug tracing if gt 0 default 0 log num qp log maximum number of QPs per HCA default is 17 max is 20 log num srq log maximum number of SRQs per HCA default is 16 max is 20 log rdmarc p
191. process If this process gets to the reset host stage and there is no path to the target from this port ib srp will Mellanox Technologies Rev 1 10 69 70 remove this 5 51 host After the scsi host is removed multipath switches to another path to this tar get from another port HCA When the failed path recovers it will be detected by the SRP daemon The SRP daemon will then request ib srp to connect to this target Once the connection is up there will be a new scsi host for this target Multipath will be executed on the devices of this host returning to the original state prior to the failed path Prerequisites Installation for RHEL4 5 Execute once Verify that the standard device mapper multipath rpm is installed If not install it from the RHEL distribution Installation for SLES10 Execute once Verify that multipath is installed If not take it from the installation you may use yast Update udev Execute once for manual activation of High Availability only Add a file to etc udev rules d you can call it 91 srp rules This file should have one line ACTION add KERNEL sd 0 9 RUN sbin multipath M m Note When SRPHA ENABLE is set to yes see Automatic Activation of High Availability below this file is created upon each boot of the driver and is deleted when the driver is unloaded Manual Activation of High Availability Initialization Execute after each bo
192. ptional Inactive Print all available information about the InfiniBand device s verbose Mellanox Technologies Rev 1 10 123 InfiniBand Fabric Diagnostic Utilities 124 Examples 1 List the names of all available InfiniBand devices gt devinfo i 2 HCAs found mthca0 mlx4 0 2 Query the device mlx4 0 and print user available information for its Port 2 gt ibv devinfo d mlx4 0 i 2 hca id mlx4 0 fw ver node guid sye mege Gute vendor id vendor perie 1615 hw ver board id PAYS JORG cim port 2 SEES S max mtu 120619 jose Hnes 11 7 ibstatus Applicable Hardware All InfiniBand devices Description 2 5 944 0000 0000 0007 3895 0000 0000 0007 3898 0 02 9 25418 _04 0140005 2 PORT ACTIVE 4 2048 4 2048 4 il 1 0 00 Displays basic information obtained from the local InfiniBand driver Output includes LID SMLID port state port physical state port width and port rate Synopsis ibstatus h device name gt lt port gt Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual Table 27 lists the various flags of the command Table 27 ibstatus Flags and Options 125 Fla Optional Default Description Mandatory If Not Specified P h Optional Print the help menu device Optional All devices Print in
193. ree routing algorithm to the guids provided in the given file one to a line cn guid file file name Set the compute nodes for the Fat Tree routing algorithm to the guids provided in the given file one to a line ids guid file file name Name of the map file with set of the IDs which will be used by Up Down routing algorithm instead of node GUIDs format guid id per line guid routing order file file name gt Set the order port guids will be routed for the MinHop and Up Down routing algorithms to the guids provided in the given file one to a line Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 90 This option causes OpenSM to configure the subnet once then exit Ports remain in the ACTIVE state S sweep interval value This option specifies th number of seconds between subnet sweeps Specifying s 0 disables sweeping Without s OpenSM defaults to a sweep interval of 10 seconds t timeout value This option specifies the time in milliseconds used for transaction timeouts Specifying t 0 disables timeouts Without t OpenSM defaults to a timeout value of 200 milli seconds maxsmps lt number gt This option specifies the number of VL15 SMP MADs allowed on the wire at any one time Specifying maxsmps 0 allows unlimited outstanding SMPs Without maxsmps OpenSM defaults to a maximum of 4 outstanding SMPs
194. restored to the default values suitable for datagram mode It is recommended to change the IPoIB mode only while the driver is down by setting the line SET to SET in the file etc infiniband openib conf and then restarting the driver Ethernet Performance Tuning When the etc init d openibd script loads the m1x4 en driver the following network stack parameters are applied net ipv4 tcp_timestamps 0 netipv4 tcp 0 net core netdev max backlog 250000 net core rmem_max 16777216 Mellanox Technologies Rev 1 10 200 net core wmem max 16777216 net core rmem default 16777216 net core wmem default 16777216 net core optmem max 16777216 net ipv4 tcp mem 16777216 16777216 16777216 net ipv4 tcp rmem 4096 87380 16777216 net ipv4 tcp wmem 4096 65536 16777216 D 3 MPI Performance Tuning To optimize bandwidth and message rate running over MVAPICH you can set tuning paramters either using the command line or in the configuration file usr mpi compiler mvapich mvapich ver etc mvapich conf Tuning Parameters in Configuration File Edit the mvapich conf file with the following lines VIADEV USE COALESCE 1 VIADEV COALESCE THRESHOLD SQ 1 VIADEV PROGRESS THRESHOLD 2 Tuning Parameters via Command Line The following command tunes MVAPICH parameters hostl usr mpi gcc mvapich lt mvapich ver gt bin mpirun rsh np 2 hostfile home lt username gt clu
195. ribution default kernels you can run scst_vdisk blockio mode to obtain good perfor mance You can also run scst_disk i e scsi pass thru mode however you have to compile scst with DSTRICT_SERIALIZING enabled but this does not yield good performance It is required to recompile the kernel to have good performance with scst_disk 2 Download and install the SCST driver a Download scst 1 0 0 tar gz from http scst sourceforge net downloads html If your distribution is RHEL 5 2 please go to step lt e gt b Untar and install scst 1 0 0 tar zxvf scst 1 0 0 tar gz cd scst 1 0 0 make amp amp make install c Save the following patch as tmp scsi tgt patch scsi tgt h2008 07 20 14 25 30 000000000 0700 scsi tgt h2008 07 20 14 25 09 000000000 0700 42 7 42 9 fendif dif LINUX VERSION CODE KERNEL VERSION 2 6 19 typedef _ 1 bool define true 1 define false 0 endif 2330 7 2332 7 Mellanox Technologies Rev 1 10 202 void scst async mcmd completed struct scst mgmt cmd mcmd int status dif LINUX VERSION CODE KERNEL VERSION 2 6 24 static inline struct page sg page struct scatterlist sg return sg gt page 2358 7 2360 7 sg gt offset offset sg gt length len endif LINUX_VERSION CODE lt KERNEL VERSION 2 5 Z4 f static inline void sg clear struct scatterlist sg d Patch scsi tgt h with tmp scsi_tgt patch
196. rovided by Mellanox BXOFED allows connecting to Fibre Channel FC tar gets on an FC fabric using an FCoE capable switch or gateway Key features include TII and pre T11 frame format Complete hardware offload of SCSI operations in pre T11 format Hardware offload of FC CRC calculations in pre T11 format Zero copy FC stack in pre T11 format e VLANs and PFC Priority flow control that is PPP The FCoE feature is based on and interacts with the Open FCoE project The m1x4 fc module is designed to replace the original fcoe module and to allow using ConnectX hardware offloads Mellanox BXOFED also includes the following open fcoe org modules libfc Used by the m1x4 fc module to handle FC logic such as fabric login and logout remote port login and logout fc ns transactions etc fcoe Implements fully in software Will load instead of mlx4 fc to support T11 frame format Works on top of standard Ethernet NICs including m1x4 en See http www open fcoe org for further information on the Open FCoE project 2 4 2 Installation To install the FCoE feature you should run the install script described in Section 2 3 with the with fc option Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 2 4 3 2 4 3 1 FCoE Basic Usage After loading the driver userspace operations should create destroy vHBAS on required Ethernet interfaces This can be done manually by issuing commands to the driver usin
197. rt con figuration will then be set for the selected device Note This utility also has a non interactive mode In this document ConnectX will be used to indicate also ConnectX 2 devices Mellanox Technologies Rev 1 10 Working With VPI 22 2 2 2 3 2 3 1 2 3 2 2 3 3 sbin connectx port config d device PCI device ID gt c conf lt portl port2 gt InfiniBand Driver The InfiniBand driver mix4 ib handles InfiniBand specific functions and plugs into the Infini Band midlayer Ethernet Driver Overview The Ethernet driver m1x4 en exposes the following ConnectX capabilities Single Dual port Fibre Channel over Ethernet FCoE Up to 16 Rx queues per port Rx steering mode Receive Core Affinity RCA Tx arbitration mode VLAN user priority off by default MSI X or INTx Adaptive interrupt moderation HW Tx Rx checksum calculation Large Send Offload 1 TCP Segmentation Offload Large Receive Offload Multi core NAPI support VLAN Tx Rx acceleration HW VLAN stripping insertion HW VLAN filtering HW multicast filtering ifconfig up down MTU changes up to 10K Ethtool support Net device statistics CXA connectors XAUI or Loading the Ethernet Driver By default the Mellanox BXOFED stack does not load m1x4 en To cause the mlx4 en module to load at driver start up set MLX4 EN LOAD yes in file etc infiniband openib conf prior to start up Alternatively i
198. s 0 oP N 16 32 64 128 256 512 1024 2048 4096 8192 16384 32768 65536 131072 262144 524288 1048576 2097152 4194304 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 640 320 160 80 40 20 10 t usec Mbytes sec 1 47 1 57 1 56 1 53 1 55 1 60 1 62 1 61 1 80 2 05 2 67 3 74 6 15 10 15 12 75 18 47 30 84 48 88 86 36 163 91 335 82 726 25 1786 35 4253 59 OUTPUT TRUNCATED 0 00 0 61 1 22 2 49 4 92 9 52 18 86 37 90 67 65 119 26 183 08 260 97 317 84 384 74 612 84 845 85 1013 28 1278 77 1447 43 1525 26 1488 90 1376 94 1119 60 940 38 Mellanox Technologies Rev 1 10 MPI 82 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 9 9 1 Quality of Service Overview Quality of Service QoS requirements stem from the realization of I O consolidation over an IB network As multiple applications and ULPs share the same fabric a means is needed to control their use of network resources Figure 4 I O Consolidation Over InfiniBand Servers Administrator QoS L3 IB Ethernet Gateway IB Fibre Block Storage Channel Gateway QoS over BXOFED for Linux is discussed in Chapter 10 OpenSM Subnet Manager The basic need is to differentiate the service levels provided to different traffic flows such that a policy can be enforced a
199. s 49 Chapter 5 RDS 51 5 1 Overview 51 5 2 RDS Configuration 51 Chapter 6 SDP 53 6 1 Overview 53 6 2 libsdp so Library 53 6 3 Configuring SDP 53 6 3 1 How to Know SDP Is Working 54 6 3 2 Monitoring and Troubleshooting Tools 54 6 4 Environment Variables 56 6 5 Converting Socket based Applications 56 6 6 BZCopy Zero Copy Send 62 6 7 Testing SDP Performance 62 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual Chapter 7 SRP 65 7 1 Overview 65 7 2 SRP Initiator 65 7 2 1 Loading SRP Initiator 65 7 2 2 Manually Establishing an SRP Connection 65 7 2 3 SRP Tools ibsrpdm and srp daemon 66 7 2 4 Automatic Discovery and Connection to Targets 68 7 2 5 Multiple Connections from Initiator IB Port to the Target 69 7 2 6 High Availability HA 69 7 2 7 Shutting Down SRP 71 Chapter 8 MPI 73 8 1 Overview 73 8 2 Prerequisites for Running MPI 73 8 2 1 SSH Configuration 73 8 3 MPI Selector Which MPI Runs 74 8 4 Compiling MPI Applications 75 8 5 OSU MVAPICH Performance 75 8 5 1 Requirements 75 8 5 2 Bandwidth Test Performance 76 8 5 3 Latency Test Performance 76 8 5 4 Intel MPI Benchmark 77 8 6 Open MPI Performance 78 8 6 1 Requirements 78 8 6 2 Bandwidth Test Performance 78 8 6 3 Latency Test Performance 79 8 6 4 Intel MPI Benchmark 80 Chapter 9 Quality of Service 83 9 1 Overview 83 9 2 QoS Architecture 84 9 3 Supported Policy 84 9 4 CMA features 85 9 5 IPoIB 85 9 6 SDP 85 9 7 RDS 85 9
200. s used Warning May result in system instability or Flash corruption if the device or another application is currently using the Flash i mage image burn verify Binary image file qq burn query Run a quick query When specified mstflint will not perform full image integrity checks during the query operation This may shorten execution time when running over slow interfaces e g I2C MTUSB 1 nofs burn Burn image in a non failsafe manner skip is burn Allow burning the firmware image without updating the invariant sector This is to ensure failsafe burning even when an invariant sector difference is detected byte mode burn write Shift address when accessing Flash internal registers May be required for burn write commands when accessing certain Flash types s ilent burn Do not print burn progress messages y es All Non interactive mode Assume the answer is to all questions no All Non interactive mode Assume the answer is no to all questions vsd lt string gt burn Write this string of up to 208 characters to VSD upon a burn command use image ps burn Burn vsd as it appears in the given image do not keep existing VSD on Flash Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 145 Table 33 mstflint Switches Sheet 2 of 2 Switch Description Commands dual image burn Ma
201. se refer to the BXOFED Installation Guide txt BXOFED for Linux is provided as a tarball tgz file The tarball contains an installation script called install p1 that performs the necessary steps to accomplish the following Discover the currently installed kernel Mellanox Technologies Mellanox BXOFED Overview 16 Uninstall any InfiniBand stacks that are part of the standard operating system distribution or another vendor s commercial stack Install the BXOFED source RPMs Build and install the BXOFED binaries Download the BXOFED X X X Y Y Y tgz file to your target Linux host If this package is to be installed on a cluster it is recommended to download it to an NFS shared directory Verify that the system has a Mellanox network adapter HCA NIC installed by ensuring that you can see ConnectX ConnectX 2 or InfiniHost entries in the display The following example shows a system with an installed Mellanox HCA host1 lspci v grep Mellanox 02 00 0 InfiniBand Mellanox Technologies MT25418 ConnectX IB DDR PCIe 2 0 2 5GT s rev a0 Use the md5sum utility to confirm the file integrity of the downloaded tarball Run the following command and compare the result to the value provided on the download page host1 md5sum BXOFED X X X Y Y Y tgz 1 3 2 Software Components BXOFED contains the following software components Network adapter drivers mthca IB only mlx4 VPI which is split into the following
202. smlid gt Optional Use lt smlid gt as the target LID for SM SA queries V ersion Optional Show version info C ca name Optional Use the specified channel adapter or router P ca port Optional Use the specified port t timeout ms Optional Override the default timeout for the solicited MADs msec Mellanox Technologies Rev 1 10 InfiniBand Fabric Diagnostic Utilities 136 Table 30 smpquery Flags and Options Optional Default ROT Mandatory If Not Specified lt gt Mandatory Supported operations nodeinfo lt addr gt nodedesc lt addr gt portinfo lt addr gt lt portnum gt switchinfo lt addr gt pkeys lt addr gt lt portnum gt sl2vl lt addr gt lt portnum gt vlarb lt addr gt lt portnum gt guids lt addr gt dest dr path lid Optional Destination s directed path LID or GUID guid gt Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 137 Examples 1 Query PortInfo by LID with port modifier gt smpquery portinfo 1 1 Port info Lid 1 port 1 INT SEN 0x0000000000000000 GidPrefix 0xfe80000000000000 id 0x0001 0x0001 0x251086a IsSM IsTrapSupported IsAutomaticMigrationSupported IsSLMappingSupported IsSystemImageGUIDsupported IsCommunicatonManagementSupported IsVendorClassSupported IsCapabilityMaskNoticeSupported IsClientReg
203. stallation Configuration Root Password e Hostname e Network Customer Center e Online Update e Service Users e Clean Up Release Notes e Hardware Configuration Mellanox Technologies Rev 1 10 Step 3 Click the Add tab in the iSCSI Initiator Overview window An iSCSI Initiator Discovery window will pop up Enter the IP Address of your iSCSI target and click Next Preparation Language License Agreement gt Disk Activation e System Analysis e Time Zone Installation e Installation Summary e Perform Installation Configuration Root Password Hostname Network Customer Center Online Update Service Users Clean Up Release Notes Help Hardware Configuration iSCSI Initiator Discovery IP Address Port 10 4 3 7 3260 No Authentication Username LJ Incoming Authentication Password Username C Outgoing Authentication Password Abort Step 4 Details of the discovered iSCSI target s will be displayed in the iSCSI Initiator Discovery window Select the target that you wish to connect to and click Connect Preparation Language y License Agreement gt Disk Activation e System Analysis e Time Zone Installation e Installation Summary Perform Installation Configuration Root Password Hostname Network Customer Center Online Update Service Users Clean Up Release Notes Hardware Configuration iSCSI
204. stem distributions Introduction to Mellanox VPI Adapters Mellanox VPI adapters which are based on Mellanox ConnectX ConnectX 2 adapter devices provide leading server and storage I O performance with flexibility to support the myriad of com munication protocols and network fabrics over a single device without sacrificing functionality when consolidating I O For example VPI enabled adapters can support Connectivity to 10 20 and 40Gb s InfiniBand switches Ethernet switches emerging Data Cen ter Ethernet switches InfiniBand to Ethernet and Fibre Channel Gateways and Ethernet to Fibre Channel gateways Fibre Channel over Ethernet FCoE and Fibre Channel over InfiniBand FCoIB Ethernet over InfiniBand EoIB A single firmware image for dual port ConnectX ConnectX 2 adapters that supports inde pendent access to different convergence networks InfiniBand Ethernet or Data Center Ethernet per port A unified application programming interface with access to communication protocols including Networking TCP IP UDP sockets Storage NFS CIFS iSCSI NFS RDMA SRP iSER Fibre Channel Clustered Storage and FCoE Clustering MPI DAPL RDS sockets and Management SNMP SMI S Communication protocol acceleration engines including networking storage clustering virtu alization and RDMA with enhanced quality of service BXOFED Package Download Tarball Note For detailed installation procedures plea
205. ster VIADEV_USE_COALESCE 1 VIADEV_COALESCE THRESHOLD SQ 1 VIADEV_PROGRESS_THRESHOLD 2 usr mpi gcc mvapich mvapich ver tests osu benchmarks osu ver osu bw The example assumes the following A cluster of at least two nodes Example host1 host2 A machine file that includes the list of machines Example host1 cat home lt username gt cluster host1 host2 host1 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 201 Appendix E SRP Target Driver The SRP Target driver is designed to work directly on top of OpenFabrics OFED software stacks bttp www openfabrics org or InfiniBand drivers in Linux kernel tree kernel org It also inter faces with Generic SCSI target mid level driver SCST http scst sourceforge net By interfacing with an SCST driver it is possiblee to work with and support a lot of IO modes on real or virtual devices in the backend 1 scst_disk interfacing with the scsi sub system to claim and export real scsi devices disks hardware raid volumes tape library as SRP luns 2 scst vdisk fileio and blockio modes This allows turning software raid volumes LVM vol umes IDE disks block devices and normal files into SRP luns 3 NULLIO mode allows measuring the performance without sending IOs to real devices E 1 Prerequisites 1 Supported distributions RHEL 5 5 1 5 2 SLES 10 sp1 and vanilla kernels gt 2 6 16 Note On dist
206. sub number lt 255 prefix length 0 9 and with value lt 32 A prefix length of 24 matches the subnet mask 255 255 255 0 A prefix length of 32 requires matching of the exact IP lt port range start port end port where port numbers are gt 0 and 65536 Note that rules are evaluated in the order of definition So the first match wins If no match is made libsdp will default to both Examples Use SDP by clients connecting to machines that belongs to subnet 192 168 1 use sdp connect 192 168 1 0 24 A family program role address port range Use SDP by ttcp when it connects to port 5001 of any machine use sdp listen ttcp 5001 Use TCP for any program with name starting with ttcp serving ports 22 to 25 use tcp server ttcp Be 22 25 Listen on both TCP and SDP by any server that listen on port 8080 use both server 8080 Mellanox Technologies Rev 1 10 SDP 58 Connect ssh through SDP and fallback to TCP to hosts on 11 4 8 port 22 use both connect 11 4 8 0 24 22 Explicit Non transparent Conversion Use explicit conversion if you need to maintain full control from your application while using SDP To configure an explicit conversion to use SDP simply recompile the application replacing PF INET or PF INET with AF INET SDP or AF INET SDP when calling the socket system call in the source code The value of AF INET SDP is defined in the file sdp socket h or you can define i
207. swO parameters set for switches port 0 qos swe parameters set for switches external ports Here s the example of typical default values for CAs and switches external ports hard coded in OpenSM initialization qos ca max vls 15 qos ca high limit O qos ca vlarb high 0 4 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 10 0 11 0 12 0 13 0 14 0 qos ca vlarb low 0 0 1 4 2 4 3 4 4 4 5 4 6 4 7 4 8 4 9 4 10 4 11 4 12 4 13 4 14 4 qos ca sl2v1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 7 qos swe max vls 15 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 113 qos swe high limit 0 qos swe vlarb high 0 4 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 10 0 11 0 12 0 13 0 14 0 qos swe vlarb low 0 0 1 4 2 4 3 4 4 4 5 4 6 4 7 4 8 4 9 4 10 4 11 4 12 4 13 4 14 4 qos swe sl2vl 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 7 VL arbitration tables both high and low are lists of VL Weight pairs Each list entry contains a VL number values from 0 14 and a weighting value values 0 255 indicating the number of 64 byte units credits which may be transmitted from that VL when its turn in the arbitration occurs A weight of 0 indicates that this entry should be skipped If a list entry is programmed for VL15 or for a VL that is not supported or is not currently configured by the port the port may either skip that entry or send from any supported VL for that entry Note that the same VLs may be listed multiple times in the High or Low prior
208. t Sections SUSE Linux Enterprise Server 10 SP2 default Floppy Failsafe SUSE Linux Enterprise Server 10 SP2 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 169 Step 15 Click Edit in the Boot Loader Settings window Section List Boot Loader Settings From Other you can manually editthe boot Section Management loader configuration files clearthe current configuration and propose a new Def configuration start from SUSE Linux Enterpris r 10 SP2 scratch or reread the Floppy Other chainloader dev fdO configuration saved on Failsafe SUSE Linux Enterprise Server 10 SP2 Image append showopts ide n your disk If you have multiple Linux systems installed YaST can try to find them and merge their menus I ag ast Er Mellanox Technologies Rev 1 10 170 Step 16 In the Optional Kernel Command Line Parameter field append the following string to the end of the line 15 mode off include a space before the string Click OK and then Finish to apply the change Section Name 4 Boot Loader Settings Section Management Use Section Name to specify the boot loader section name The section name must be unique t Section Editor Section Settings Selecting Do not Section Name verify Filesystem SUSE Linux Enterprise Server 10 SP2 befor
209. t CPU utilization Server CPU utilization Separates the global and test specific parameters r 1 1 The request size sent and how many bytes requested back Note that the run example above produced the following results e Client CPU utilization is 5 61 percent of client CPU Server CPU utilization is 6 79 percent of server CPU Latency is 25 11 microseconds Latency is calculated as follows 0 5 1 Transaction rate per sec 1 000 000 one way average latency in usec Step 6 To end the test shut down the Netperf server host1 pkill netserver Mellanox Technologies Rev 1 10 IPoIB 38 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 4 4 1 4 2 EoIB Introduction The Ethernet over IB EoIB mlx4 vnic module is a network interface implementation over Infini Band EoIB encapsulates Layer 2 datagrams over an InfiniBand Datagram UD transport service The InfiniBand UD datagrams encapsulates the entire Ethernet L2 datagram and its payload Figure 2 EoIB in the OSI Model Application Transport Internet Link EoIB In order to perform this operation the module performs an address translation from Ethernet layer 2 MAC addresses 48 bits long to InfiniBand layer 2 addresses made of LID GID and QPN This translation is done by the module and is totally invisible to the OS and user In this regard EoIB differs from IPoIB which ex
210. t inline define AF INET SDP 27 define PF INET SDP AF INET SDP You can compile and execute the following very simple TCP application that has been converted explicitly to SDP Compilation gec sdp server c sdp server gcc sdp client c o sdp client Usage Server hostl sdp server Client hostl sdp client server IPoIB addr gt Example Server hostl sdp server accepted connection from 15 2 2 42 48710 read 2048 bytes end of test host1 Client host2 sdp client 15 2 2 43 connected to 15 2 2 43 22222 sent 2048 bytes host2 sdp client c Code usage sdp client lt ip addr gt ail Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual include lt stdio h gt include lt stdlib h gt include lt stdint h gt include lt unistd h gt include lt string h gt include lt sys types h gt include lt sys socket h gt include lt netinet in h gt include lt arpa inet h gt define DEF PORT 22222 define AF_INET_SDP 27 define PF INET SDP AF INET SDP define TXBUFSZ 2048 uint8 ttx buffer TXBUFSZ int main int argc char argv if arge lt 2 printf Usage sdp client lt ip_addr gt n exit EXIT FAILURE j int sd socket PF INET SDP SOCK STREAM 0 1 54 lt 0 perror socket failed exit EXIT FAILURE struct sockaddr in to_addr sin family AF INET sin port htons DEF PORT
211. t is highly recommended to create and assign a swap partition Swap partitions on your system are listed in the main window with the type Linux Swap An assigned swap partition has the mount point swap You can assign more than one swap partition if desired Do you wantto change this scsi Create Edit Delete Resize EVMS J RAID v fi Crypt File v Expert Step 14 Select the Expert tab and click Booting Preparation Language License Agreement Disk Activation System Analysis Time Zone Installation gt Installation Summary w Perform Installation Configuration e Root Password e Hostname e Network e Customer Center e Online Update e Service e Users e Clean Up e Release Notes e Hardware Configuration Show Release Notes a Installation Settings Click any headline to make changes or use the Change menu below Overview Expert Keyboard Layout English US Partitioning Create swap partition dev sdal 502 0 MB Create root partition dev sda2 7 5 GB with reiserfs Add On Products No add on product selected for installation Software SUSE Linux Enterprise Server 10 SP2 Server Base System KDE Desktop Environment for Server C C Compiler and Tools X Window System Size of Packagesto Install 1 6 GB Booting Boot Loader Type GRUB Location dev sda2 boo
212. t performance on the PCI Express interface the adapter card should be installed in an x8 slot with the following BIOS configuration parameters Max Read Req the maximum read request size is 512 or higher e MaxPayloadSize the maximum payload size is 128 or higher Note A Max Read Req of 128 and or installing the card in an x4 slot will significantly limit bandwidth To obtain the current setting for Max Read enter setpci d 15b3 68 w To obtain the PCI Express slot link width and speed enter setpci d 15b3 72 1 If the output is neither 81 nor 82 card then the card is NOT installed in an x8 PCI Express slot 2 The least significant digit indicates the link speed for PCI Express Gen 1 2 5 GT s 2 for PCI Express Gen 2 5 GT s Note If you are running InfiniBand at QDR 40Gb s 4X IB ports you must run PCI Express Gen 2 InfiniBand Performance Troubleshooting InfiniBand IB performance depends on the health of IB link s and on the IB card type IB link speed 10Gb s or SDR 20Gb s or DDR 40Gb s or QDR also affects performance Note A latency sensitive application should take into account that each switch on the path adds 200nsec at SDR and 150nsec for DDR To check the IB link speed enter ibstat Check the value indicated after the Rate string 10 indicates SDR 20 indicates DDR and 40 indicates QDR 2 Check that the link has NO symbol errors since these errors result
213. t sockaddr_in my_addr sin family AF INET sin port htons DEF PORT sin addr s addr INADDR ANY int retbind bind sd struct sockaddr amp my_addr sizeof my_addr if retbind lt 0 perror bind failed exit EXIT FAILURE j int retlisten listen sd 5 backlog if retlisten 0 perror listen failed exit EXIT FAILURE j accept the client connection struct sockaddr in client addr socklen t client addr len sizeof client_addr int cd accept sd struct sockaddr amp client addr amp client addr len if ed 0 perror accept failed exit EXIT FAILURE j printf accepted connection from s u n inet ntoa client addr sin addr ntohs client addr sin port ssize t nr read cd rx buffer RXBUFSZ if nr 0 perror read failed exit EXIT FAILURE else if nr 0 printf socket was closed by remote host n printf read zd bytes n nr printf end of test n close cd close sd return 0 61 Mellanox Technologies Rev 1 10 SDP 62 6 6 BZCopy Zero Copy Send BZCOPY mode is only effective for large block transfers By setting the sys parameter sdp zcopy thresh to a non zero value a non standard SDP speedup is enabled Messages longer than sdp zcopy thresh bytes in length cause the user space buffer to be pinned and the data to be sent directly from the original buffer This res
214. t the Mellanox Firmware Tools MFT package has been installed on the client machine hostl mst start 1 mst status The device name will be of the form dev mst mt lt dev_id gt _pci _cr0 conf0 Use this device name to obtain the MAC address via a query command flint d lt MST DEVICE NAME gt q Example with ConnectX EN as the network adapter device hostl flint d dev mst mt25448 pci cr0 q Image type ConnectX FW Version 2 6 0 Rom Info type GPXE version 1 3 0 devid 25448 Device ID 25448 Chip Revision AO Description Portl Port2 MACs 0002c90000bb 0002c90000bc Board ID n a MT 0920110004 VSD n a PSID MT 0920110004 Assuming that ConnectX EN PXE is connected via Port 1 then the MAC address is 00 02 c9 00 00 bb Extracting the MAC Address Method II The six bytes of MAC address can be captured from the display upon the boot of the ConnectX device session as shown in the figure below 11 Connei gPXE 0 9 5 Open Source Boot Firmware open Link up Waiting for link up on 0 Placing MAC Addresses in etc dhepd conf The following is an excerpt of a etc dhcpd conf example file showing the format of represent ing a client machine for the DHCP server running on a Linux machine host host1 next server 11 4 3 7 filename pxelinux 0 fixed address 11 4 3 130 hardware ethernet 00 02 c9 00 00 bb j Mellanox Technologies Rev 1 10 178
215. tall Mellanox BXOFED remember to choose srpt y install pl E 2 How to run A On an SRP Target machine 1 Please refer to SCST s README for loading scst driver and its dev_handlers drivers scst_disk scst vdisk block or file IO mode nullio Note Regardless of the mode you always need to have lun 0 in any group s device list Then you can have any lun number following lun 0 it is not required to have the lun numbers in ascending order except that the first lun must always be 0 Setting SRPT LOAD yes in etc infiniband openib conf is not enough as it only causes the load ing ofthe ib srpt module but it does not load scst and its dev handlers Example 1 Working with VDISK BLOCKIO mode Using md0 device sda and cciss c1d0 a modprobe scst b modprobe scst_vdisk echo open vdisk0 dev md0 BLOCKIO gt proc scsi_tgt vdisk vdisk Mellanox Technologies Rev 1 10 204 oe a echo open vdisk1 dev sda BLOCKIO gt proc scsi tgt vdisk vdisk echo open vdisk2 dev cciss c1d0 BLOCKIO gt proc scsi_tgt vdisk vdisk echo add vdisk0 0 gt proc scsi_tgt groups Default devices echo add vdisk1 1 gt proc scsi_tgt groups Default devices echo add vdisk2 2 gt proc scsi_tgt groups Default devices Example 2 Working with real back end scsi disks in scsi pass thru mode modprobe scst modprobe scst disk cat proc scsi tgt scsi tgt ibstor00 cat proc scsi tgt scsi tgt
216. target IP and port number ULPs might also provide QoS Class The CMA then creates Service ID for the ULP and passes this ID and optional QoS Class in the PR MPR request The resulting PR MPR is used for configuring the connection QP nN PathRecord and MultiPathRecord Enhancement for QoS As mentioned above the PathRecord and MultiPathRecord attributes are enhanced to carry the Ser vice ID which is a 64bit value A new field QoS Class is also provided A new capability bit describes the SM QoS support in the SA class port info This approach pro vides an easy migration path for existing access layer and ULPs by not introducing new set of PR MPR attributes Supported Policy The QoS policy which is specified in a stand alone file is divided into the following four subsec tions I Port Group A set of CAs Routers or Switches that share the same settings A port group might be a partition defined by the partition manager policy list of GUIDs or list of port names based on NodeDescrip tion Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 9 4 9 5 9 6 9 7 I Fabric Setup Defines how the SL2VL and VLArb tables should be setup Note In BXOFED this part of the policy is ignored SL2VL and VLArb tables should be config ured in the OpenSM options file opensm opts III QoS Levels Definition This section defines the possible sets of parameters for QoS that a client mi
217. te gt ibstatus mlx4 0 1 Infiniband device mlx4_0 port 1 status default gid fe80 0000 0000 0000 0000 0000 9289 3895 base lid 0x3 sm lid 0x3 state 2 INIT phys state 5 LinkUp rate 20 Gb sec 4X DDR gt ibportstate C mlx4 0 3 1 query PortInfo Port info Lid 3 port 1 EinkState ae ees Initialize PhysLinkState LinkUp LinkWidthSupported 1X or 4X LinkWidthEnabled 1X or 4X LinkWidthA ctive 4X LinkSpeedSupported 2 5 Gbps or 5 0 Gbps LinkSpeedEnabled 2 5 Gbps or 5 0 Gbps LinkSpeedA ctive 5 0 Gbps Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 129 2 Query the status of two channel adapters using directed paths gt ibportstate C mlx4_0 D 0 1 PortInfo Port info DR path slid 65535 dlid 65535 0 port 1 linkStaie Initialize PhysLinkState LinkUp LinkWidthSupported or 4X Link WidthEnabled 1X or 4X LinkWidthA ctive 4X LinkSpeedSupported 2 5 Gbps or 5 0 Gbps LinkSpeedEnabled 2 5 Gbps or 5 0 Gbps LinkSpeedActive 5 0 Gbps gt ibportstate C mthca0 D 0 1 PortInfo Port info DR path slid 65535 dlid 65535 0 port 1 linkStaie e Down PhysLinksState Polling LinkWidthSupported 1X or 4X LinkWidthEnabled
218. tem Analysis gt Time Zone Installation e Installation Summary e Perform Installation Configuration Hostname Root Password Network Customer Center Online Update Service Users Clean Up Release Notes Hardware Configuration Egg Clock and Time Zone Region Time Zone Canada Central and South America Russia Asia Australia Africa Pacific Global Etc Alaska Aleutian Arizona Central East Indiana Hawaii Indiana Starke Michigan Mountain Pacific Samoa Hardware Clock Set To Abort 2 Date 07 52 06 24 03 2008 Finish Mellanox Technologies Rev 1 10 187 188 Step 10 In the Installation Settings window click Partitioning to get Suggested Partitioning window Preparation v Language v License Agreement v Disk Activation v v Installation Settings System Analysis Click any headline to make changes or use the Change menu below Time Zone Overview Expert Installation gt Installation Summary Keyboard Layout e Perform Installation English US Configuration e Hostname Partitioning eop pasaar Create boot partition dev sda1 70 5 MB with ext2 e Network Create swap partition dev sda2 502 0 MB Customer Center Create root partition dev sda3 7 4 GB with reiserfs e Online Update e Service Software e Users e Clean Up SUSE Linux Enterprise Server 10 e Release
219. terfaces via the vconfig command In order to create interfaces with VLAN refer to Chapter 4 3 5 VLAN Configuration 4 5 Retrieving EoIB information 4 5 1 4 vnic info To retrieve information about EoIB interfaces use the script mlx4 vnic info This script gives detailed information about a specific vNic or all EoIB vNic interfaces If network administered vNics are enabled this script can also be used to discover the available BridgeXs from the host side To do this simply run mlx4 vnic info grep SYSTEM GUID or mlx4 vnic info grep SYSTEM NAME To get the full vNic information simply type mlx4 vnic info or mlx4 vnic info eth10 A typical output for a single interface looks something like NETDEV NAME eth10 NETDEV_LINK up NETDEV_OPEN yes NETDEV_QSTOP no Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual GW PORT ID SL GW QPN GW LID IB PORT IB LOG LINK IB PHY LINK IB MTU MTU RX RINGS NUM RX RINGS LIN SW RSS SW RSS SIZE TX RINGS NUM TX RINGS ACT NDO TSS NDO TSS SIZE PROMISC MCAST MCAST MASK NO BXM LRO ENABLED LRO NUM NAPI ENABLED VNIC PER PORT VNIC NUM PKEY PKEY INDEX QPN LID MAC VID VLAN USED GID GW PORT NAME SYSTEM NAME SYSTEM GUID GW EPORT 0 0 0x800000 0x0004 mlx4 0 1 LinkUp Active 2048 1500 Oxffff 0x0000 0x80060 0x0003 00 00 00 00 02 00 0x000 0 fe 80 00 00 00 00 00 00 00 02 c9 03 00 00 22 85
220. the OpenSM options file default location var cache opensm opensm opts IIT QoS Levels denoted by qos levels Each QoS Level defines Service Level SL and a few optional fields MTU limit Rate limit PKey Packet lifetime When path s search is performed it is done with regards to restriction that these QoS Level param eters impose One QoS level that is mandatory to define is a DEFAULT QoS level It is applied to a PR MPR query that does not match any existing match rule Similar to any other QoS Level it can also be explicitly referred by any match rule IV QoS Matching Rules denoted by qos match rules Each PathRecord MultiPathRecord query that OpenSM receives is matched against the set of matching rules Rules are scanned in order of appearance in the QoS policy file such as the first match takes precedence Each rule has a name of QoS level that will be applied to the matching query A default QoS level is applied to a query that did not match any rule Queries can be matched by Source port group whether a source port is a member of a specified group Destination port group same as above only for destination port PKey QoS class Service ID To match a certain matching rule PR MPR query has to match ALL the rule s criteria However not all the fields of the PR MPR query have to appear in the matching rule For instance if the rule has a single criterion Service ID it will match
221. the TCP IP Port Number that the protocol connects to The default port number for RDS is 0x48CA which makes a default Service ID 0x00000000010648CA 85 Mellanox Technologies Rev 1 10 86 Quality of Service 9 8 9 9 SRP The current SRP implementation uses its own CM callbacks not CMA So SRP fills in the Ser vice ID in the PR MPR by itself and use that information in setting up the QP SRP Service ID is defined by the SRP target I O Controller it also complies with IBTA Service ID rules The Service ID is reported by the I O Controller in the ServiceEntries DMA attribute and should be used in the PR MPR if the SA reports its ability to handle QoS PR MPRs OpenSM Features The QoS related functionality that is provided by OpenSM the Subnet Manager described in Chapter 10 can be split into two main parts I Fabric Setup During fabric initialization the Subnet Manager parses the policy and apply its settings to the dis covered fabric elements II PR MPR Query Handling OpenSM enforces the provided policy on client request The overall flow for such requests is first the request is matched against the defined match rules such that the target QoS Level definition is found Given the QoS Level a path s search is performed with the given restrictions imposed by that level Note QoS in OpenSM is described in detail in Chapter 10 Rev 1 10 Mellanox Technologies Mellanox BXOFED Stack f
222. the interface enter the ifconfig command with the following items The appropriate IB interface 160 161 etc The IP address that you want to assign to the interface The netmask keyword The subnet mask that you want to assign to the interface The following example shows how to configure an IB interface host1 ifconfig ib0 11 4 3 175 netmask 255 255 0 0 Step 2 Optional Verify the configuration by entering the i config command with the appro priate interface identifier ib argument The following example shows how to verify the configuration host1 ifconfig ibO b0 Link encap UNSPEC HWaddr 80 00 04 04 FE 80 00 00 00 00 00 00 00 00 00 00 inet addr 11 4 3 175 Beast 11 4 255 255 Mask 255 255 0 0 UP BROADCAST MULTICAST MTU 65520 Metric 1 RX packets 0 errors 0 dropped 0 overruns 0 frame 0 TX packets 0 errors 0 dropped 0 overruns 0 carrier 0 collisions 0 txqueuelen 128 RX bytes 0 0 0 b TX bytes 0 0 0 b Step 3 Repeat Step and Step 2 on the remaining interface s Subinterfaces You can create subinterfaces for a primary IPoIB interface to provide traffic isolation Each such subinterface also called a child interface has a different IP and network addresses from the pri mary parent interface The default Partition Key PKey ff ff applies to the primary parent interface This section describes how to Create a subinterface Section 3 4 1 Remove a subinterface Section 3 4 2 Creating a Subinterf
223. tiator ext value for the relevant path If you use srp daemon with n flag it automatically assigns initiator ext values according to this convention For example id_ext 200500A0B81146A1 ioc_guid 0002c90200402bec dgid fe800000000000000002c90200402bed pkey ffff service 1d 200500a0b81146al nitiator ext ed2b400002c90200 7 2 6 Notes It is recommended to use the n flag for all srp daemon invocations 2 ibsrpdm does not have a corresponding option 3 srp daemon sh always uses the n option whether invoked manually by the user or automati cally at startup by setting SRPHA ENABLE to yes High Availability HA Overview High Availability works using the Device Mapper DM multipath and the SRP daemon Each ini tiator is connected to the same target from several ports HCAs The DM multipath is responsible for joining together different paths to the same target and for fail over between paths when one of them goes offline Multipath will be executed on newly joined SCSI devices Each initiator should execute several instances of the SRP daemon one for each port At startup each SRP daemon detects the SRP Targets in the fabric and sends requests to the ib srp module to connect to each of them These SRP daemons also detect targets that subsequently join the fabric and send the ib srp module requests to connect to them as well Operation When a path from port1 to a target fails the ib srp module starts an error recovery
224. transmission burst Such configuration would suilt VL that needs low latency and uses small MTU when transmitting packets Rest of VLs are defined as low priority VLs with different weights while VL4 is effectively turned off 10 6 8 Deployment Example Figure 6 shows an example of an InfiniBand subnet that has been configured by a QoS manager to provide different service levels for various ULPs Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 114 Figure 6 Example QoS Deployment on InfiniBand Subnet Traffic class Partition A Service level 0 Policy min 40 Ap A Banc App Server Traffic class SDP Service level 2 Policy min 20 BW App B Server ZA Service SL1 m Virtual Server Access Points Traffic class SRP Service Level 1 Traffic class IPoIB Policy min 30 BW Service Level 3 Policy min 10 BW i 0 App Server Virtual Server App B Server 10 7 QoS Configuration Examples The following are examples of QoS configuration for different cluster deployments Each example provides the QoS level assignment and their administration via OpenSM configuration files 10 7 1 Typical HPC Example MPI and Lustre Assignment of QoS Levels MPI Separate from I O load Min BW of 7096 Storage Control Lustre MDS Low latency Storage Data Lustre OST Min BW 30 Administration MPlis assigned an SL via the command line 1 mpirun s1 0
225. ts If you are not i Size Type Mount Start Used By Label Devil familiar with the 8 0 GB IET VIRTUAL DISK 0 1045 scsi concepts of hard disk Ndev sda 70 5 8 partitions and how to use them you might wantto go back and select automatic partitioning F Linux native Ext2 boo K 0 8 7 4 GB F Linux native Reiser 73 1045 Please note that nothing will be written to your hard disk until you confirm the entire installation You have not assigned a swap partition There is nothing wrong with that but in the last installation in most cases it is highly recommended to create and assign a swap partition dialog Until that point Swap partitions on your system are listed in the main window with the you can safely abort type Linux Swap An assigned swap partition hasthe mount point swap the installation You can assign more than one swap partition if desired For LVM setup using a non LVM root device and a non LVM swap device is recommended Other than the root and swap devices you should have partitions managed by LVM Do you wantto change this The table to the right shows the current partitions all your Create Edit Delete Resize hard disks B iwm EVMS J RAID v fi Crypt File v Expert v Hard disks are Step 14 Select the Expert tab and click Booting Preparation
226. ty can be invoked in one of two ways 1 The mpi selector menu command This command is a simple menu based program that allows the selection of the system wide MPI usu ally only settable by root and a per user MPI selection It also shows what the current selections are This command is recommended for all users 2 The mpi selector command This command is a CLI equivalent of the mpi selector menu allowing for the same functionality as mpi selector menu but without the interactive menus and prompts It is suitable for scripting Compiling MPI Applications Note A valid Fortran compiler must be present in order to build the MVAPICH MPI stack and tests The following compilers are supported by Mellanox BXOFED s MVAPICH and Open MPI pack ages Intel and PGI The install script prompts the user to choose the compiler with which to install the MVAPICH and Open MPI RPMs Note that more than one compiler can be selected simultaneously if desired Compiling MVAPICH Applications Please refer to http mvapich cse ohio state edu support mvapich user guide html To review the default configuration of the installation check the default configuration file usr mpi lt compiler gt mvapich lt mvapich ver gt etc mvapich conf Compiling Open MPI Applications Please refer to http www open mpi org faq category mpi apps 85 OSU MVAPICH Performance 8 5 1 Requirements At least two nodes Example host1 host2 Mac
227. ulticast packet The table is organized by MLID Network Interface Card NIC A network adapter card that plugs into the PCI Express slot and provides one or more ports to an Ethernet network Standby Subnet Manager Mellanox Technologies Rev 1 10 210 A Subnet Manager that is currently quiescent and not in the role of a Master Subnet Manager by agency of the master SM See Subnet Manager Subnet Administrator SA An application normally part of the Subnet Manager that implements the interface for querying and manipulating subnet management data Subnet Manager SM One of several entities involved in the configuration and control of the subnet Unicast Linear Forwarding Tables LFT A table that exists in every switch providing the port through which packets should be sent to each LID Virtual Protocol Interconnet VPI A Mellanox Technologies technology that allows Mellanox channel adapter devices ConnectX to simultaneously connect to an InfiniBand subnet and a 10GigE subnet each subnet connects to one of the adpater ports Rev 1 10 Mellanox Technologies
228. ults in less CPU usage and on many sys tems much higher bandwidth Note that the default value of sdp zcopy thresh is 64KB but is may be too low for some sys tems You will need to experiment with your hardware to find the best value 6 7 Testing SDP Performance This section describes how to verify SDP performance by running the Bandwidth BW test and the Latency test These tests are described in detail at the following URL http www netperf org netperf training Netperf html To verify SDP performance perform the following steps Step 1 Download Netperf from the following URL http www netperf org netperf NetperfPage html Step 2 Compile Netperf by following the instructions at http www netperf org netperf NetperfPage html Step 3 Create Libsdp conf configuration file hostl cat gt HOME libsdp conf lt lt EOF gt use sdp server gt use sdp client gt EOF Step 4 Start the Netperf server such that you force SDP to be used instead of TCP hostl LD PRELOAD libsdp so LIBSDP CONFIG FILE HOME libsdp conf netserver Starting netserver at port 12865 Starting netserver at hostname 0 0 0 0 port 12865 and family AF UNSPEC hostili Step 5 Run the Netperf client such that you force SDP to be used instead of TCP The default test is the Bandwidth test host2 LD PRELOAD libsdp so LIBSDP CONFIG FILE HOME libsdp conf netperf H 11 4 17 6 t TCP STREAM c C m 65536 TCP STREAM TEST from 0 0 0 0 0 0 0 0 port 0 AF INET
229. umber of gather scatter entries per I O default 12 Manually Establishing an SRP Connection The following steps describe how to manually load an SRP connection between the Initiator and an SRP Target Section 7 2 4 explains how to do this automatically Make sure that the ib module is loaded the SRP Initiator is reachable by the SRP Target and that an SM is running To establish a connection with an SRP Target and create an SRP SCSI device for that target under dev use the following command Mellanox Technologies Rev 1 10 66 echo n id ext GUID value ioc guid GUID value dgid port GID value N pkey ffff service id service 0 value gt sys class infiniband srp srp mthca hca number port number add target See Section 7 2 3 for instructions on how the parameters in this echo command may be obtained Notes Execution of the above echo command may take some time The SM must be running while the command executes tis possible to include additional parameters in the echo command max per lun Default 63 max sect short for max sectors sets the request size of a command io class Default 0x100 as in rev 16A of the specification In rev 10 the default was Oxff00 initiator ext Please refer to Section 9 Multiple Connections To list the new SCSI devices that have been added by the echo command you may use either of the following two methods Execute f
230. vel driver implementation for the ConnectX adapters designed by Mellanox Technologies The ConnectX can operate as an InfiniBand adapter as an Ethernet NIC or as a Fibre Channel HBA To accommodate the supported configurations the driver is split into four modules mlx4 core Handles low level functions like device initialization and firmware commands processing Also controls resource allocation so that the InfiniBand and Ethernet functions can share the device without interfering with each other mlx4 ib Handles InfiniBand specific functions and plugs into the InfiniBand midlayer Mellanox Technologies Rev 1 10 Mellanox BXOFED Overview 18 mlx4 en A 10GigE driver under drivers net mlx4 that handles Ethernet specific functions and plugs into the netdev mid layer mlx4 fc Handles the FCoE functions using ConnectX Fibre Channel hardware offloads mlx4 vnic Handles the EoIB functions using ConnectX Ethernet hardware offloads 1 4 5 Mid layer Core Core services include management interface MAD connection manager CM interface and Subnet Administrator SA interface The stack includes components for both user mode and ker nel applications The core services run in the kernel and expose an interface to user mode for verbs CM and management 1 4 4 Open FCoE The FCoE feature is based on and interacts with the Open FCoE project BXOFED includes the following open fcoe org modules libfc and fcoe See Section 2 4
231. with a specific Service ID any pkey 0x0OABC 6 match PR MPR query with a specific PKey srp target port guid 0x1234 5 SRP when SRP Target 15 located on a specified IB port GUID any target port guid OXOABC OxFFFFF 6 match any PR MPR query with a specific target port GUID end qos ulps Mellanox Technologies Mellanox BXOFED Stack for Linux User s Manual 111 Similar to the advanced policy definition matching of PR MPR queries is done in order of appear ance in the QoS policy file such as the first match takes precedence except for the default rule which is applied only if the query didn t match any other rule All other sections of the QoS policy file take precedence over the qos ulps section That is if a policy file has both qos match rules and qos ulps sections then any query is matched first against the rules in the qos match rules section and only if there was no match the query is matched against the rules in qos ulps section Note that some of these match rules may overlap so in order to use the simple QoS definition effec tively it is important to understand how each of the ULPs is matched 10 6 6 1IPoIB IPoIB query is matched by PKey or by destination GID in which case this is the GID of the multi cast group that OpenSM creates for each IPoIB partition Default PKey for IPoIB partition is Ox 7fff so the following three match rules are equivalent Ipoib lt SL gt ipoib pke
232. y 0x7fff lt SL gt any pkey 0 7 lt SL gt 10 6 6 2SDP SDP PR query is matched by Service ID The Service ID for SDP is 0x000000000001PPPP where PPPP are 4 hex digits holding the remote TCP IP Port Number to connect to The following two match rules are equivalent sdp DSSL any service id 0x0000000000010000 0x000000000001ffff lt SL gt 10 6 6 3 RDS Similar to SDP RDS PR query is matched by Service ID The Service ID for RDS is 0x000000000106PPPP where PPPP are 4 hex digits holding the remote TCP IP Port Number to connect to Default port number for RDS is 0x48CA which makes a default Service ID 0x00000000010648CA The following two match rules are equivalent rds lt SL gt any service id 0x00000000010648CA SL 10 6 6 4 iSER Similar to RDS iSER query is matched by Service ID where the the Service ID is also 0x000000000106PPPP Default port number for iSER is 0xOCBC which makes a default Service ID 0x0000000001060CBC The following two match rules are equivalent Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 112 iser lt SL gt any service id 0x0000000001060CBC lt SL gt 10 6 6 5SRP Service ID for SRP varies from storage vendor to vendor thus SRP query is matched by the target IB port GUID The following two match rules are equivalent srp target port guid 0x1234 lt SL gt any target port guid 0x1234 SL Note that any of the above ULPs might contain target
233. y are leaf switches Switches of the same rank should have the same number of ports in each UP going port group Switches of the same rank should have the same number of ports in each DOWN going port group All the CAs have to be at the same tree level rank If the root guid file is provided the topology doesn t have to be pure fat tree and it should only comply with the following rules Tree rank should be between two and eight inclusively 1 Ports that are connected to the same remote switch are referenced as port group 101 Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 102 All the Compute Nodes have to be at the same tree level rank Note that non compute node CAs are allowed here to be at different tree ranks Topologies that do not comply cause a fallback to min hop routing Note that this can also occur on link failures which cause the topology to no longer be a fat tree Note that although fat tree algorithm supports trees with non integer CBB ratio the routing will not be as balanced as in case of integer CBB ratio In addition to this although the algorithm allows leaf switches to have any number of CAs the closer the tree is to be fully populated the more effective the shift communication pattern will be In general even if the root list is pro vided the closer the topology to a pure and symmetrical fat tree the more optimal the routing will be
234. y file First matched rule takes precedence qos match rules matching by single criteria QoS class qos match rule use by QoS class qos class 7 9 11 Name of qos level to apply to the matching PR MPR qos level name WholeSet end qos match rule show matching by destination group and service id qos match rule use Storage targets destination Storage service id 0x10000000000001 0x10000000000008 0x10000000000FFF qos level name WholeSet end qos match rule qos match rule source Storage use match by source group only qos level name DEFAULT end qos match rule qos match rule use match by all parameters qos class 7 9 11 source Virtual Servers destination Storage service id 0x0000000000010000 0x000000000001FFFF pkey OxOF00 0xOFFF qos level name WholeSet end qos match rule end qos match rules 10 6 6 Simple QoS Policy Details and Examples Simple QoS policy match rules are tailored for matching ULPs or some application on top of a ULP PR MPR requests This section has a list of per ULP or per application match rules and the SL that should be enforced on the matched PR MPR query Match rules include Default match rule that 1s applied to PR MPR query that didn t match any of the other match rules Mellanox Technologies Rev 1 10 OpenSM Subnet Manager 110 SDP SDP application with a specific target TCP IP port range SRP with a specific target IB port GUID RDS iSER

BXOFED User‟s Manual

Contents

Download Pdf Manuals

Related Search

Related Contents