Mellanox ScalableSHMEM User Manual
Contents
1. Rev 2 1 2 5 1 Enabling MXM for ScalableSHMEM Jobs MXM is activated automatically in ScalabeSHMEM for jobs with Number of Elements PE higher or equal to 128 Toenable MXM for SHMEM jobs for any PE Add the following MCA parameter to the shmemrun command line mca spml ikrit np number For additional MXM tuning information please refer to the MellanoX Messaging Library README file found in the Mellanox website 2 6 Running ScalableSHMEM with FCA The Mellanox Fabric Collective Accelerator FCA is a unique solution for offloading collective operations from the Message Passing Interface MPI or ScalableSHMEM process onto Mellanox InfiniBand managed switch CPUs As a system wide solution FCA utilizes intelligence on Mella nox InfiniBand switches Unified Fabric Manager and MPI nodes without requiring additional hardware The FCA manager creates a topology based collective tree and orchestrates an efficient collective operation using the switch based CPUs on the MPI ScalableSHMEM nodes FCA accelerates MPI ScalableSHMEM collective operation performance by up to 100 times pro viding a reduction in the overall job runtime Implementation is simple and transparent during the job runtime FCA is disabled by default and must be configured prior to using it from the Scal ableSHMEM gt To enable FCA by default in the ScalableSHMEM 1 Edit the opt mellanox openshmem 2 1 etc openmpi mca params conf file2. ScalableSHMEM contains support for environment module system http mod 4 ules sf net The modules configuration file can be found at ye opt mellanox openshmem 2 1 etc shmem modulefile 2 3 1 Basic ScalableSHMEM Job Run Example Tolaunch ScalableSHMEM application run opt mellanox openshmem 2 1 bin shmemrun np 2 mca mpi paffinity alone 1 bynode display map hostfile myhostfile opt mellanox openshmem 2 1 bin shmem osu latency The example above shows how to run 2 copies of shmem osu latency program on hosts specified in the myhostfile file Mellanox Technologies 7 Rev 2 1 Installing and Configuring ScalableSHMEM ScalableSHMEM package is compiled with SLURM support However if you do not use SLURM job scheduler on your system you can disable it for SHMEM job as following opt mellanox openshmem 2 1 bin shmemrun np 2 mca grpcomm pmi mca pubsub pmi mca ess pmi opt mellanox openshmem 2 1 bin shmem osu latency 2 4 ScalableSHMEM Tunable Parameters ScalableSHMEM uses Modular Component Architecture MCA parameters to provide a way to tune your runtime environment Each parameter corresponds to a specific function The following are parameters that you can change their values to change the application s the function memheap controls memory allocation policy and thresholds scoll controls ScalableSHMEM collective API threshold and algorithms spml controls ScalableSHMEM point to point transport
3. 2 Set the scoll_fca_enable parameter to 1 scoll fca_enable 1 To enable FCA in the snmemrun command line add the following mca scoll fca_enable 1 To disable FCA mca scoll fca enable 0 mca coll fca enable 0 For more details on FCA installation and configuration please refer to the FCA User Manual found in the Mellanox website Mellanox Technologies 9 Rev 2 1 Installing and Configuring ScalableSHMEM 2 7 Configuring ScalableSHMEM with External Profiling Tools TAU ScalableSHMEM supports external Tuning and Profiling tool TAU For further information please refer to http www cs uoregon edu Research tau home php 2 7 1 Using TAU with OpenSHMEM 2 7 1 1 Building a PDT Toolkit 1 Download the PDT Toolkit wget nc http tau uoregon edu pdt releases pdtoolkit 3 17 tar gz tar xzf pdtoolkit 3 17 tar gz 2 Configure and build the PDT toolkit ciel exe ttererel lite LT PDT INST PWD configure prefix usr local make install 2 7 1 2 Building a TAU Toolkit 1 Download the TAU Toolkit wget nc http www cs uoregon edu research paracomp tau tauprofile dist tau latest tar gz cele Gait iesu distet ieee oU 2 Configure and build the TAU toolkit cd tau latest TAU_SRC PWD patch pl i opt mellanox openshmem 2 1 share openshmem tau_openshmem patch OSHMEM INST opt mellanox openshmem 2 1 TAU INST TAU SRC install configure prefix TAU INST shmem tag oshmem cc gcc pdt PDT INST PROF
4. 7 23 Running ScalableSHMEM Application 7 2 3 1 Basic ScalableSHMEM Job Run Example 0 0 0 ee 7 2 4 ScalableSHMEM Tunable Parameters 8 2 5 Running ScalableSHMEM with MXM 8 2 5 1 Enabling MXM for ScalableSHMEM Jobs sssseseee ee neces 9 2 6 Running ScalableSHMEM with FCA 9 2 7 Configuring ScalableSHMEM with External Profiling Tools TAU 10 2 7 1 Using TAU with OpenSHMEM 00 eect m 10 Appendix A Performance Optimization 0 cee cece cee cece enn n nnn n nn nl AT Configunng Hugepages 1420 2 m uc ei Reb ates ne rr ne Geld Sua a AER UAR En ged 11 A 2 Tuning MTU Size to the Recommended Value sssseeeee ee 12 A 3 HPC Applications on Intel Sandy Bridge Machines 0 0 0 cece eee eens 13 Mellanox Technologies 3 J Rev 2 1 Shared Memory Access Overview 1 Shared Memory Access Overview The Shared Memory Access SHMEM routines provide low latency high bandwidth communi cation for use in highly parallel scalable programs The routines in the SHMEM Application Pro gramming Interface API provide a programming model for exchanging data between cooperating parallel processes The SHMEM API can be used either alone or in combination with MPI routines in the same parallel progr
5. logic and thresholds atomic controls ScalableSHMEM atomic operations logic and thresholds shmem controls general ScalableSHMEM API behavior To display ScalableSHMEM parameters 1 2 Print all available parameters Run opt mellanox openshmem 2 1 bin shmem info a Print ScalableSHMEM specific parameters Run opt mellanox openshmem 2 1 bin shmem info param shmem all opt mellanox openshmem 2 1 bin shmem info param memheap all opt mellanox openshmem 2 1 bin shmem info param scoll all opt mellanox openshmem 2 1 bin shmem info param spml all opt mellanox openshmem 2 1 bin shmem info param atomic all 2 5 Running ScalableSHMEM with MXM MellanoX Messaging MXM library provides enhancements to parallel communication libraries by fully utilizing the underlying networking infrastructure provided by Mellanox HCA switch hardware This includes a variety of enhancements that take advantage of Mellanox networking hardware including Multiple transport support including RC XRC and UD Proper management of HCA resources and memory structures Efficient memory registration One sided communication semantics Connection management Receive side tag matching Intra node shared memory communication These enhancements significantly increase the scalability and performance of message communi cations in the network alleviating bottlenecks within the parallel communication libraries 8 Mellanox Technologies
6. utilizing all node sockets When installing MLNX OFED 1 8 an automatic work around is activated upon Sandy Bridge machine detection and the following message is printed in the job s standard output device m1x4 Sandy Bridge CPU was detected To disable MOFED 1 8 Sandy Bridge NUMA related workaround Setthe SHELL environment variable before launching HPC application Run export MLX4 STALL CQ POLL 0 shmemrun lt gt or shmemrun x MLX4 STALL CQ POLL 0 lt other params gt Mellanox Technologies 13 J
7. ILEPARAM use ropt ISOSHMEM INST include mpp shmemlib OSHMEM INST lib shmemlibrary lshmem lpmi make install The patch is required to define a profiling API that is not part of an official openshmem org standard 10 Mellanox Technologies Rev 2 1 Appendix A Performance Optimization A 1 Configuring Hugepages Hugepages is a feature applicable to users using MLNX OFED v1 5 3 3 0 0 gt Hugepages can be allocated using the proc sys vm nr_hugepages entry or by using the sysctl command To view the current setting using the proc entry cat proc sys vm nr hugepages 0 Toview the current setting using the sysctl command sysctl vm nr hugepages vm nr hugepages 0 Tosetthe number of huge pages using proc entry echo 1024 gt proc sys vm nr hugepages Tosetthe number of hugepages using sysctl man sysctl w vm nr hugepages 1024 vm nr hugepages 1024 To allocate all the hugepages needed you might need to reboot your system since the hugepages requires large areas of contiguous physical memory In time physical memory may be mapped and allocated to pages thus the physical memory can become fragmented If the hugepages are allocated early in the boot process fragmentation 1s unlikely to have occurred It is recommended that the etc sysct1 conf file be used to allocate hugepages at boot time For example to allocate 1024 hugepages at boot time add the line below to th
8. Mellanox TECHNOLOGIES Mellanox ScalableSHMEM User Manual Rev 2 1 www mellanox com Rev 2 1 NOTE THIS HARDWARE SOFTWARE OR TEST SUITE PRODUCT PRODUCT S AND ITS RELATED DOCUMENTATION ARE PROVIDED BY MELLANOX TECHNOLOGIES AS IS WITH ALL FAULTS OF ANY KIND AND SOLELY FOR THE PURPOSE OF AIDING THE CUSTOMER IN TESTING APPLICATIONS THAT USE THE PRODUCTS IN DESIGNATED SOLUTIONS THE CUSTOMER S MANUFACTURING TEST ENVIRONMENT HAS NOT MET THE STANDARDS SET BY MELLANOX TECHNOLOGIES TO FULLY QUALIFY THE PRODUCTO S AND OR THE SYSTEM USING IT THEREFORE MELLANOX TECHNOLOGIES CANNOT AND DOES NOT GUARANTEE OR WARRANT THAT THE PRODUCTS WILL OPERATE WITH THE HIGHEST QUALITY ANY EXPRESS OR IMPLIED WARRANTIES INCLUDING BUT NOT LIMITED TO THE IMPLIED WARRANTIES OF MERCHANTABILITY FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT ARE DISCLAIMED IN NO EVENT SHALL MELLANOX BE LIABLE TO CUSTOMER OR ANY THIRD PARTIES FOR ANY DIRECT INDIRECT SPECIAL EXEMPLARY OR CONSEQUENTIAL DAMAGES OF ANY KIND INCLUDING BUT NOT LIMITED TO PAYMENT FOR 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 LIABILITY OR TORT INCLUDING NEGLIGENCE OR OTHERWISE ARISING IN ANY WAY FROM THE USE OF THE PRODUCT S AND RELATED DOCUMENTATION EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE Mellanox TECHNOLOGIES Mellanox Te
9. am The SHMEM parallel programming library is an easy to use programming model which uses highly efficient one sided communication APIs to provide an intuitive global view interface to shared or distributed memory systems SHMEM s capabilities provide an excellent low level inter face for PGAS applications A SHMEM program is of a single program multiple data SPMD style All the SHMEM pro cesses referred as processing elements PEs start simultaneously and run the same program Commonly the PEs perform computation on their own sub domains of the larger problem and periodically communicate with other PEs to exchange information on which the next communica tion phase depends The SHMEM routines minimize the overhead associated with data transfer requests maximize bandwidth and minimize data latency the period of time that starts when a PE initiates a transfer of data and ends when a PE can use the data SHMEM routines support remote data transfer through put operations data transfer to a different PE get operations data transfer from a different PE and remote pointers allowing direct references to data objects owned by another PE Additional supported operations are collective broadcast and reduction barrier synchronization and atomic memory operations An atomic memory operation is an atomic read and update opera tion such as a fetch and increment on a remote or local data object SHMEM libraries implement act
10. chnologies Mellanox Technologies Ltd 350 Oakmead Parkway Suite 100 Beit Mellanox Sunnyvale CA 94085 PO Box 586 U S A Yokneam 20692 www mellanox com Israel Tel 408 970 3400 Tel 972 4 909 7200 Fax 408 970 3403 972 74 723 7200 Fax 972 4 959 3245 Copyright 2012 Mellanox Technologies All rights reserved Mellanox BridgeX ConnectX CORE Direct InfiniBlast InfiniBridge InfiniHost InfiniRISC InfiniScale amp InfiniPCI PhyX SwitchX UFM Virtual Protocol Interconnect and Voltaire are registered trademarks of Mellanox Technologies Ltd FabricIT MLNX OS Unbreakable Link UFM and Unified Fabric Manager are trademarks of Mellanox Technologies Ltd All other trademarks are property of their respective owners 2 Mellanox Technologies Document Number 3708 Rev 2 1 Table of Contents Fable Of Contents ie ee Se Beer sella ies v Roe re ee eue eie e i aee Chapter 1 Shared Memory Access Overview ccc cec cece csc re ee ceccscseeee 4 1 1 Mellanox ScalableSHMEM 4 1 2 ScalableSHMEM Supported Platforms and Operating Systems 5 Chapter 2 Installing and Configuring ScalablSHMEM cece cccccecccecees 2 1 Installing ScalableSHMEM 6 22 Compiling ScalableSHMEM Application
11. e sysctl conf file vm nr hugepages 1024 Mellanox Technologies 11 Rev 2 1 Installing and Configuring ScalableSHMEM A 2 Tuning MTU Size to the Recommended Value The procedures described below apply to user using MLNX OFED 1 5 3 3 0 0 only gt When using MLNX_OFED 1 5 3 3 0 0 it is recommended to change the MTU to 4k Whereas in MLNX OFED 1 8 the MTU is already set by default to 4k To check the current MTU support of an InfiniBand port use the smpquery tool smpquery D PortInfo 0 1 grep i mtu If the MtuCap value is lower than 4K enable it to 4K Assuming the firmware is configured to support 4K MTU the actual MTU capability is further limited by the mlx4 driver parameter To further tune it 1 Set the set 4k mtu mlx4 driver parameter to 1 on all the cluster machines For instance echo options mlx4 core set 4k mtu 1 gt gt etc modprobe d mofed conf 2 Restart openibd by running the following command service openibd restart Tocheck whether the parameter was accepted run cat sys module mlx4 core parameters set 4k mtu To check whether the port was brought up with 4K MTU this time use the smpquery tool again 12 Mellanox Technologies Rev 2 1 A 3 HPC Applications on Intel Sandy Bridge Machines This section applies to user using MLNX OFED 1 8 only 5 Intel Sandy Bridge machines have NUMA hardware related limitation which affects performance of HPC jobs
12. ive messaging The sending of data involves only one CPU where the source processor puts the data into the memory of the destination processor Likewise a pro cessor can read data from another processor s memory without interrupting the remote CPU The remote processor is unaware that its memory has been read or written unless the programmer implements a mechanism to accomplish this 1 1 Mellanox ScalableSHMEM The ScalableSHMEM programming library is a one side communications library that supports a unique set of parallel programming features including point to point and collective routines syn chronizations atomic operations and a shared memory paradigm used between the processes ofa parallel programming application Mellanox ScalableSHMEM is based on the API defined by the OpenSHMEM org consortium The library works with the OpenFabrics RDMA for Linux stack OFED and also has the ability to uti 4 Mellanox Technologies Rev 2 1 lize MellanoX Messaging libraries MXM as well as Mellanox Fabric Collective Accelerations FCA providing an unprecedented level of scalability for SHMEM programs running over Infini Band The latest ScalableSHMEM software can be downloaded from the Mellanox website 1 2 ScalableSHMEM Supported Platforms and Operating Systems Platforms x86 64 Operating Systems RHEL 5 5 5 6 5 7 6 0 6 1 62 SLESIO SP4 SLESII SP1 OFED MLNX OFED 1 5 3 3 0 0 or higher The following pac
13. kages must me installed separately on all cluster nodes when using MLNX OFED 1 5 3 3 0 0 Mellanox MXM Mellanox FCA KNEM KMEM is a Linux module enabling high performance intra node SHMEM communication for large messages KNEM is not started automatically after the installation process To start it manually run as root modprobe knem For information on how to starts KNEM automatically after system boot please refer to KNEM doc umentation Mellanox OFED 1 8 includes all the packages above which are installed under MXM opt mellanox mxm 4 e FCA opt mellanox fca KNEM opt knem 0 9 7mlnx1 If you have installed OFED 1 8 you do not need to download and install them Mellanox Technologies 5 J Rev 2 1 Installing and Configuring ScalableSHMEM 2 installing and Configuring ScalableSHMEM 2 1 Installing ScalableSHMEM MLNX OFED v1 8 includes ScalableSHMEM 2 1 which is installed under opt mellanox openshmem 2 1 If you have installed OFED 1 8 you do not need to download and install ScalableSH aa MEM The ScalableSHMEM package is an rpm file and should be installed on all cluster nodes It is built to support the SLURM job scheduler by utilizing a PMI API The procedure below is required only if you have installed MLNX OFED 1 5 3 3 0 0 If you have installed OFED 1 8 you do not need to download and install ScalableSH Ad MEM To install ScalableSHMEM perform the following step
14. s 1 Login as root Run rpm ihv openshmem 2 1 20624 x86 64 rpm nodeps 2 Update the OPENIB kernel parameter in the etc modprobe conf file on all cluster nodes echo options mlx4 core log num mtt 24 gt gt etc modprobe conf 3 Restart the openibd driver etc init d openibd restart 6 Mellanox Technologies J Rev 2 1 2 2 Compiling ScalableSHMEM Application The ScalableSHMEM package contains a shmemcc utility which is used as a compiler command To compile ScalableSHMEM application 1 Save the code example below as a file called example c include lt stdio h gt include lt stdlib h gt include lt shmem h gt int main int argc char argv int my pe num pe shmem init my pe ny pe num pe num pes printf Hello World from process d of d n my pe num pes exit 0 2 Compile the example with the SHMEM C wrapper compiler opt mellanox openshmem 2 1 bin shmemcc o example exe example c 2 3 Running ScalableSHMEM Application The ScalableSHMEM framework contains the shmemrun utility which launches the executable from a service node to compute nodes This utility accepts the same command line parameters as mpirun from the OpenMPI package For further information please refer to OpenMPI MCA parameters documentation at http www open mpi org faq category running Run shmemrun help to obtain ScalableSHMEM job launcher runtime parameters
Download Pdf Manuals
Related Search