142
szerkesztés
Módosítások
a
Basically the SGE is a scheduler, which divides the resources, computers into resource partitions. These are called queues. A queue can’t be larger than a physical resource; it can’t expand its borders. SGE registers a waiting list for the resources managed by itself, to which the posted computing tasks are directed. The scheduler searches for the resource defined by the description of the task and starts it. The task-resource coupling depends on the ability of the resources and the parameters of the tasks. In case the resources are overloaded, the tasks have to wait while the requested processor and memory becomes available.== Acknowledgement in publications ==
The detailed documentation of the SGE can be found '''We acknowledge [http:PRACE/KIFÜ] for awarding us access to resource based in Hungary at [Budapest/docs.oracle.com/cd/E24901_01Debrecen/doc.62Pécs/e21976.pdf hereSzeged].'''
SGE version on all HPC sites: '''We acknowledge KIFÜ for awarding us access to resource based in Hungary at [http:Budapest/Debrecen/gridscheduler.sourceforge.netPécs/documentation.html Open Grid Scheduler (OGS/GE 2011Szeged].11p1)]'''
The most simple commands The most simple SGE command is the display of the cluster data: qhostA possible outcome of this command can be: HOSTNAME ARCH NCPU LOAD MEMTOT MEMUSE SWAPTO SWAPUS ------------------------------------------------------------------------------- global - - - - - - - uv linux-x64 1152 900.56 6057.9G 132.4G 0.0 0.0 The first two columns define the names and types of the computers, which are in the cluster. The NCPU column shows the number of the available processor cores. LOAD shows the computer’s load for the moment (this value equals with the value demonstrated by the uptime UNIX command). The rest of the cells are: overall physical memory, the actual used memory, the available swap-memory, and the used swap. The global line marks all the information in total regarding the cluster.We can have a look at the available queue-s with the following command: qconf -sqlOne probable outcome of the command: test.q uv.q To get more info about the state of the system use qstat -fIt shows which jobs run in which queues, and you can also get detailed info about the queues themselves (state, environment). The command can be used without the -f switch too, but it is less informative, since in this case only the jobs’ states will appear. The command’s outcome: queuename qtype resv/used/tot. load_avg arch states --------------------------------------------------------------------------------- test.q@uv BIP 0/1/30 800.15 linux-x64 905 1.00000 PI_SEQ_TES stefan r 06/04/2011 09:12:14 1 --------------------------------------------------------------------------------- uv.q@uv BIP 0/802/1110 800.15 linux-x64 The first column of this table shows the name of the row, the second column marks the type (B-batch, I-interactive, C-checkpointing, P-parallel environment, E-error state). The third part of the column shows how many jobs can be run at the same time in the row. All in all, these values fit to the number of overall processor cores in the system. The second item of the column shows the free compartments at the moment.If a running (scheduled) job is to be found in the queue, it is directly next to the name of the row, like the recent "PI_SEQ_TES", which runs in the test.q row. The tasks waiting for the resources, because it is overwhelmed or the preliminary conditions are not prompt, appear behind the sum row, listed as pending jobs. For example: queuename qtype resv/used/tot. load_avg arch states --------------------------------------------------------------------------------- test.q@uv BIP 0/0/30 600.42 linux-x64 --------------------------------------------------------------------------------- uv.q@uv BIP 0/598/1110 600.42 linux-x64 ############################################################################ - PENDING JOBS - PENDING JOBS - PENDING JOBS - PENDING JOBS - PENDING JOBS ############################################################################ 905 0.00000 PI_SEQ_TES stefan qw 06/04/2011 09:12:04 1 Each task is given an identifier, which is a number (a job ID, or j_id), this is followed by the job’s priority (0 in both cases), then the job’s name, and the user who posted the job, and the qw marks, that the job is waiting for the queue. Finally the date of the registration for the waiting queue is nextWhen a job finishes running, this is created: jobname.ojobnumber in our actual catalog, which contains the error messages and stapled outputs created by the program..Job submissionBack then, the SGE scheduler was designed to be able to operate different types of architectures. That’s why you can’t post binary files directly, only scripts, like the qsub script.shcommand. The script describes the task, the main parameters of it, and its running. For example in the following script, the describedhostname.sh task: #!/bin/sh #$ -N HOSTNAME /bin/hostname can be posted with the following command: qsub hostname.shThe scripts can be used for separating the different binaries: #!/bin/sh case `uname` in SunOS) ./pi_sun FreeBSD) ./pi_bsd esacWith the following command, we can define the queue where the scheduler puts the job:: qsub -q serial.q range.shThe command qsub can be issued with a number of different switches, which are gathered in the following table:ParameterPossible exampleResult-N name-N FlowThe job will appear under this name in the queue.-cwd-cwdThe output and the error files will appear in this actual catalog.-S shell-S /bin/tcshThe shell in which the scripts run.-j {y|n}-j yJoining the error and the output in one file.-r {y|n}-r yAfter a restart, should the job restart too (from the beginning).-M e-mail-M stefan@niif.huScheduler information will be sent to this address about the job.-l-l h_cpu=0:15:0Chooses a queue for the job where 15 minutes of CPU time could be ensured. (hour:minute:second)-l-l h_vmem=1GChooses a computer for the job where 1 GB memory is available. In the case of parallel jobs its value is extended with the required number of slots. If this parameter is not given, the default setting will be the number of the maximum memory cores set up in the computers.-l-l inConsuming resources, complex request. (This will be defined in the documentation written for the system administrators)-binding-binding linear:4Chooses 4 CPU cores on the worker node-on and assignes in a fix way. Further information. here.-l-l exclusive=trueDemand of exclusive task execution (another job will not be scheduled on the chosen computers). It can be used in the following sites[[Category: Szeged, Budapest és Debrecen.-P-P niifiChooses a HPC project. This command will list the available HPC projects: qconf -sprjl-R -R yResource reservation. This will cause that bigger parallel jobs will get higher priority. qsub command arguments can be added to the ~/.sge_request file. If this file exists then it will be added to the qsub arument list.Sometimes we want to delete a job before its running. For this you can use the qdel job_idcommand. qdel 903The example deletes the job number 903. qdel -f 903It can delete the running jobs immediately.For pending and then continuing jobs, use qmod {-s,-us}. qmod -s 903 qmod -us 903 The previous one suspends the running of number 903 (SIGSTOP), while the latter one allows (SIGCONT).If there is a need to change the features (resource requirements) of a job put into the waiting list, it can be done with the command: qalter qalter -l h_cpu=0:12:0 903 The previous command alternates the hard-CPU requirements of the job number 903 (h_cpu) and changes it to 12 minutes. The switches of the qalter command are mainly overlap the ones of the qsub command.In a special case, we have to execute the same task, but on different data. These tasks are the array jobs. With SGE we can upload several jobs to the waiting. For example in the pi task shown in previous chapter, it can be posted multiple times, with different parameters, with the following script:array.sh #!/bin/sh #$ -N PI_ARRAY_TEST ./pi_gcc `expr $SGE_TASK_ID \* 100000` The SGE_TASK_ID is an internal integer used by the SGE, which created values for each running job. The interval can be set up when posting the block: qsub -t 1-7 array.sh meaning that the array.sh program will run in seven issues, and the SGE_TASK_ID will have the value of 1, 2, ..., 7 in every running issue. The qstat -f shows how the block tasks are split: --------------------------------------------------------------------------------- test.q@uv BIP 0/0/30 8.15 linux-x64 --------------------------------------------------------------------------------- uv.q@uv BIP 0/7/1110 8.15 linux-x64 907 1.00000 PI_ARRAY_T stefan r 06/04/2011 10:34:14 1 1 907 0.50000 PI_ARRAY_T stefan t 06/04/2011 10:34:14 1 2 907 0.33333 PI_ARRAY_T stefan t 06/04/2011 10:34:14 1 3 907 0.25000 PI_ARRAY_T stefan t 06/04/2011 10:34:14 1 4 907 0.20000 PI_ARRAY_T stefan t 06/04/2011 10:34:14 1 5 907 0.16667 PI_ARRAY_T stefan t 06/04/2011 10:34:14 1 6 907 0.14286 PI_ARRAY_T stefan t 06/04/2011 10:34:14 1 7 It is clear, that behind the tasks there are their array index with which we can refer to the components to the task. For example, in the case of block tasks, there is a possibility to delete particular parts of the block. If we want to delete the subtasks from 5-7 of the previous task, the command qdel -f 907.5-7 will delete chosen components, but leaves the tasks 907.1-4 intact.The result of the running is seven individual files, with seven different running solutions:It can happen; that the task placed in the queue won’t start. This case the: qstat -j job_id command will show the detailed scheduling information, containing which running parameters are unfulfilled by the task.The priority of the different tasks only means the gradiation listed in the pending jobs. The scheduler will analyze the tasks in this order. Since it requires the reservation of resources, it is not sure, that the tasks will run exactly the same order.If we wonder why a certain job won’t start, here’s how you can get information: qalter -w v job_idOne possible outcome Job 53505 cannot run in queue "szeged.q" because it is not contained in its hard queue list (-q) Job 53505 (-l NONE) cannot run in queue "cn46.szeged.hpc.niif.hu" because exclusive resource (exclusive) is already in use Job 53505 (-l NONE) cannot run in queue "cn48.szeged.hpc.niif.hu" because exclusive resource (exclusive) is already in use Job 53505 cannot run in PE "mpi" because it only offers 0 slots verification: no suitable queuesYou can check with this command where the jobs are running: qhost -j -qHOSTNAME ARCH NCPU LOAD MEMTOT MEMUSE SWAPTO SWAPUS ------------------------------------------------------------------------------- global - - - - - - - cn01 linux-x64 48 41.43 126.0G 3.0G 0.0 0.0 serial.q BI 0/42/48 120087 0.15501 run.sh roczei r 09/23/2012 14:25:51 MASTER 22 120087 0.15501 run.sh roczei r 09/23/2012 15:02:21 MASTER 78 120087 0.15501 run.sh roczei r 10/01/2012 07:58:21 MASTER 143 120087 0.15501 run.sh roczei r 10/01/2012 08:28:51 MASTER 144 120087 0.15501 run.sh roczei r 10/04/2012 17:41:51 MASTER 158 120340 0.13970 pwhg.sh roczei r 09/24/2012 23:24:51 MASTER 3 120340 0.13970 pwhg.sh roczei r 09/24/2012 23:24:51 MASTER 5 120340 0.13970 pwhg.sh roczei r 09/24/2012 23:24:51 MASTER 19 120340 0.13970 pwhg.sh roczei r 09/24/2012 23:24:51 MASTER 23 120340 0.13970 pwhg.sh roczei r 09/24/2012 23:24:51 MASTER 31 120340 0.13970 pwhg.sh roczei r 09/24/2012 23:24:51 MASTER 33 120340 0.13970 pwhg.sh roczei r 09/26/2012 13:42:51 MASTER 113 120340 0.13970 pwhg.sh roczei r 10/01/2012 07:43:06 MASTER 186 120340 0.13970 pwhg.sh roczei r 10/01/2012 07:58:36 MASTER 187 ... Queue types parallel.q - for paralel jobs (jobs can run maximum 31 days)serial.q - for serial jobs (jobs can run maximum 31 days)test.q - test queue, the job will be killed after 2 hoursGetting information on the waiting line’s status:qstat -g c CUSTER QUEUE CQLOAD USED RES AVAIL TOTAL aoACDS cdsuE -------------------------------------------------------------------------------- parallel.q 0.91 460 0 44 504 0 0 serial.q 0.84 200 0 40 240 0 0 test.q 0.00 0 0 24 24 0 0Running PVM jobTo run the previously shown and translated gexample application, we need the following task-describing gexample.sh script: #!/bin/sh #$ -N GEXAMPLE ./gexample << EOL 30 5 EOL We can submit this with the following command: qsub -pe pvm 5 gexample.sh The -pe pvm 5 command will tell to the SGE to create a PVM parallel computer machine with 5 virtual processors, and run the application in this. uv.q@uv BIP 0/5/1110 5.15 linux-x64 908 1.00000 GEXAMPLE stefan r 06/04/2011 13:05:14 5 Also note that after the running two output files were created: one containing an attached standard error and standard output (GEXAMPLE.o908), another describing the working method of the (GEXAMLE.po908). The latter one is mainly for finding errors.Running MPI jobsAll computers are set up with several installations of the MPI system: vendor-specific MPI implementations, and MPICH system too. The default setup is the vendor-specific MPI.Running in the MPI environment is similar to the PVM environment. Let’s have a look at the example shown in the previous chapter connectivity. A very simple task which tests the MPI tasks’internal communication. Use the following connectivity.sh script to run it: #!/bin/sh #$ -N CONNECTIVITY mpirun -np $NSLOTS ./connectivity Here, the $NLOTS variable indicates that how many processors should be used in the MPI environment. This equals with that number what we have reuired for the parallel environment. The job can be submitted with the following command: qsub -pe mpi 20 connectivity.sh With this command we instruct the scheduler to create a parallel MPI environment containing 20 processors, and reserve space for it in one of the queues. Once the space is available, the job starts: uv.q@uv BIP 0/20/1110 20.30 linux-x64 910 1.00000 CONNECTOVI stefan r 06/04/2011 14:03:14 20 Running the program will result in two files: the first one (CONNECTIVITY.o910) is the overlap of the result of the already run program standard output and standard error, while the second one (CONNECTIVITY.po910) is for the follow-up of the operation of the parallel environment. If the running is successful, this file is empty. The command -pe mpi 20 can be given in the script too with the directive #$ -pe mpi 20Important notes: you should use mpirun.sge by SGI MPT on the Debrecen supercomputer when you run a job under SGE. This can automatic parse which machines have been selected by SGE.This way you can check that you are using SGI MPT or not: DEBRECEN[service0] ~ (1)$ type mpirun mpirun is hashed (/opt/nce/packages/global/sgi/mpt/2.04/bin/mpirun) DEBRECEN[service0] ~ (0)$ type mpirun.sge mpirun.sge is hashed (/opt/nce/packages/global/sgi/mpt/2.04/bin/mpirun.sge) DEBRECEN[service0] ~ (0)$ You should use mpirun binary directly if you are using SHF3 environment or you would like to use a more complex MPI run. However, you need to parse the SGE's PE_HOSTFILE environment variable in this case.Running OpenMP jobsThere are applications that either use the solutions of the operation system for multi-threaded program execution, or use a special library designed for this, like OpenMP. These applications have to be instructed how many threads they can use. The matrix multiplication algorithm presented in the previous chapter can be described with the following omp_mm.sh script #!/bin/sh #$ -N OPENMP_MM ./omp_mm it can be submitted with this command which will use 6 threads qsub -pe openmp 6 omp_mm.shCheckpointing supportAt the moment the system doesn’t support any automatic checkpointing/restarting mechanism. If it is need, the application has to take care of it.
→Acknowledgement in publications
== User Guide to obtain a digital certificate ==
<code>
gsissh -p 2222 prace-login.budapest.hpcsc.niif.hu
</code>
<code>
globus-url-copy file://task/myfile.c gsiftp://prace-login.budapestsc.hpcniif.hu/home/taskprace/pr1hrocz/myfile.c
</code>
* -stripe Use this parameter to initiate a “striped” GridFTP transfer that uses more than one node at the source and destination. As multiple nodes contribute to the transfer, each using its own network interface, a larger amount of the network bandwidth can be consumed than with a single system. Thus, at least for “big” (> 100 MB) files, striping can considerably improve performance.
==Usage of the Sun SLURM scheduler ==Website: http://slurm.schedmd.com The schedule of the HPCs are CPU hour based. This means that the available core hours are divided between users on a monthly basis. All UNIX users are connected to one or more account. This scheduler account is connected to an HPC project and a UNIX group. HPC jobs can only be sent by using one of the accounts. The core hours are calculated by the multiplication of wall time (time spent running the job) and the CPU cores requested.For example reserving 2 nodes (48 cpu cores) at the NIIFI SC for 30 minutes gives 48 * 30 = 1440 core minutes = 24 core hours. Core hours are measured between the start and and the end of the jobs. '''It is very important to be sure the application maximally uses the allocated resources. An empty or non-optimal job will consume allocated core time very fast. If the account run out of the allocated time, no new jobs can be submitted until the beginning of the next accounting period. Account limits are regenerated the beginning of each month.''' Information about an account can be listed with the following command:<code> sbalance</code> ==== Example ====After executing the command, the following table shows up for Bob. The user can access, and run jobs by using two different accounts (foobar, barfoo). He can see his name marked with * in the table. He shares both accounts with alice (Account column). The consumed core hours for the users are displayed in the second row (Usage), and the consumption for the jobs ran as the account is displayed in the 4th row. The last two row defines the allocated maximum time (Account limit), and the time available for the machine (Available). <pre>Scheduler Account Balance---------- ----------- + ---------------- ----------- + ------------- -----------User Usage | Account Usage | Account Limit Available (CPU hrs)---------- ----------- + ---------------- ----------- + ------------- -----------alice 0 | foobar 0 | 0 0bob * 0 | foobar 0 | 0 0 bob * 7 | barfoo 7 | 1,000 993alice 0 | barfoo 7 | 1,000 993</pre> === Estimating core time ===Before production runs, it is advised to have a core time estimate. The following command can be used for getting estimate: <code> sestimate -N NODES -t WALLTIME</code>where <code>NODES</code> are the number of nodes to be reserved, <code>WALLTIME</code> is the maximal time spent running the job. '''It is important to provide the core time to be reserved most precisely, because the scheduler queue the jobs based on this value. Generally, a job with shorter core time will be run sooner. It is advised to check the time used to run the job after completion with <code>sacct</code> command.''' ==== Example ==== Alice want to reserve 2 days 10 hours and 2 nodes, she checks, if she have enough time on her account.<pre>sestimate -N 2 -t 2-10:00:00 Estimated CPU hours: 2784</pre>Unfortunately, she couldn't afford to run this job. === Status information === Jobs in the queue can be listed with <code>squeue</code> command, the status of the cluster can be retrieved with the <code>sinfo</code> command. All jobs sent will get a JOBID. The properties of a job can be retrieved by using this id. Status of a running or waiting job:<code> scontrol show job JOBID </code> All jobs will be inserted into an accounting database. The properties of the completed jobs can be retrieved from this database. Detailed statistics can be viewed by using this command:<code> sacct -l -j JOBID</code> Memory used can be retrieved by using<code> smemory JOBID</code> Disk usage can be retrieved by this command:<code> sdisk JOBID</code> ==== Example ==== There are 3 jobs in the queue. The first is an array job which is waiting for resources (PENDING). The second is an MPI job running on 4 nodes for 25 minutes now. The third is an OMP run running on one node, just started. The NAME of the jobs can be freely given, it is advised to use short, informative names. <pre> squeue -l Wed Oct 16 08:30:07 2013 JOBID PARTITION NAME USER STATE TIME TIMELIMIT NODES NODELIST(REASON)591_[1-96] normal array alice PENDING 0:00 30:00 1 (None) 589 normal mpi bob RUNNING 25:55 2:00:00 4 cn[05-08] 590 normal omp alice RUNNING 0:25 1:00:00 1 cn09</pre> This two-node batch job had a typical load of 10GB virtual, and 6.5GB RSS memory per node. <pre> smemory 430 MaxVMSize MaxVMSizeNode AveVMSize MaxRSS MaxRSSNode AveRSS---------- -------------- ---------- ---------- ---------- ----------10271792K cn06 10271792K 6544524K cn06 6544524K 10085152K cn07 10085152K 6538492K cn07 6534876K </pre> ==== Checking jobs ==== It is important to be sure the application fully uses the core time reserved. A running application can be monitored with the following command:<code> sjobcheck JOBID</code> ===== Example ===== This job runs on 4 nodes. The LOAD group provides information about the general load of the machine, this is more or less equal to the number of cores. The CPU group gives you information about the exact usage. Ideally, values of the <code>User</code> column are over 90. If the value is below that, there is a problem with the application, or it is not optimal, and the run should be ended. This example job fully using ("maxing out") the available resources. <pre>Hostname LOAD CPU Gexec CPUs (Procs/Total) [ 1, 5, 15min] [ User, Nice, System, Idle, Wio]cn08 24 ( 25/ 529) [ 24.83, 24.84, 20.98] [ 99.8, 0.0, 0.2, 0.0, 0.0] OFFcn07 24 ( 25/ 529) [ 24.93, 24.88, 20.98] [ 99.8, 0.0, 0.2, 0.0, 0.0] OFFcn06 24 ( 25/ 529) [ 25.00, 24.90, 20.97] [ 99.9, 0.0, 0.1, 0.0, 0.0] OFFcn05 24 ( 25/ 544) [ 25.11, 24.96, 20.97] [ 99.8, 0.0, 0.2, 0.0, 0.0] OFF</pre> ==== Checking licenses ==== The used and available licenses can be retrieved with this command: <code> slicenses</code> ==== Checking downtime ==== In downtime periods, the scheduler doesn't start new jobs, but jobs can be sent. The periods can be retrieved by using the following command: <code> sreservations</code> === Running jobs === Running applications in the HPC can be done in batch mode. This means all runs must have a job script containing the resources and commands needed. The parameters of the scheduler (resource definitions) can be given with the <code>#SBATCH</code> directive. Comparison of the schedulers, and the directives available at slurm are available at this [http://slurm.schedmd.com/rosetta.pdf table]. ==== Obligatory parameters ====The following parameters are obligatory to provide:<pre>#!/bin/bash#SBATCH -A ACCOUNT#SBATCH --job-name=NAME#SBATCH --time=TIME</pre> where <code>ACCOUNT</code> is the name of the account to use (available accounts can be retrieved with the <code>sbalance</code> command), <code>NAME</code> is the short name of the job, <code>TIME</code> is the maximum walltime using <code>DD-HH:MM:SS</code> syntax. Acceptable time formats include "minutes", "minutes:seconds", "hours:minutes:seconds", "days-hours", "days-hours:minutes" and "days-hours:minutes:seconds". The following command submit jobs:<code> sbatch jobscript.sh</code> If the submission was successful, the following is outputted:<pre>Submitted batch job JOBID</pre>where <code>JOBID</code> is the unique id of the job The following commmand cancels the job:<code> scancel JOBID</code> ==== Job queues ==== There are two separate queue (partition) available in the HPC, the <code>test</code> queue and the <code>prod</code> queue. Tha latter is for the production runs, the former is for testing purposes. In the test queue, 1 node can be allocated for the maximum of half hours, The default queue is <code>prod</code>. Test partition can be chosen with the following directive:<pre>#SBATCH --partition=test</pre> ==== Quality of Service (QoS) ==== There is an option for submitting low priority jobs. These jobs can be interrupted by any normal priority job at any time, but only the half of the time is billed to the account. Interrupted jobs will be automatically queued again. Therefore it is important to only run jobs that can be interrupted at any time, periodically saves their states (checkpoint) and can restart quickly.The default QoS is <code>normal</code>, non-interruptable. The following directive choses low priority:<pre>#SBATCH --qos=lowpri</pre> ==== Memory settings ==== 1000 MB memory is allocated for 1 CPU core by default, more can be allocated with the following directive:<pre>#SBATCH --mem-per-cpu=MEMORY</pre>where <code>MEMORY</code> is given in MB. The maximum memory/core at NIIFI SC is 2600 MB. ==== Email notification ====Sending mail when the status of the job change (start, stop, error):<pre>#SBATCH --mail-type=ALL#SBATCH --mail-user=EMAIL</pre>where <code>EMAIL</code> is the e-mail to notify. ==== Array jobs ====Array jobs are needed, when multiple one threaded (serial) jobs are to be sent (with different data). Slurm stores unique id of the instances in the <code>SLURM_ARRAY_TASK_ID</code> enviromnemt variable. It is possible to seperate threads of the array job by retrieving these ids. Output of the threads are written into <code>slurm-SLURM_ARRAY_JOB_ID-SLURM_ARRAY_TASK_ID.out</code> files. The scheduler uploads outputs tightly. It is useful to use multiply threads for a CPU core. [http://slurm.schedmd.com/job_array.html More on this topic] ===== Example =====Alice user submits 96 serial job for a maximum of 24 hour run. on the expenses of 'foobar' account. The <code>#SBATCH --array=1-96</code> directive indicates, that it is an array job. The application can be run with the <code>srun</code> command. This is a shell script in this example.<pre>#!/bin/bash#SBATCH -A foobar#SBATCH --time=24:00:00#SBATCH --job-name=array#SBATCH --array=1-96srun envtest.sh</pre> ==== MPI jobs ====Using MPI jobs, the number of MPI processes running on a node is to be given (<code>#SBATCH --ntasks-per-node=</code>). The most frequent case is to provide the number of CPU cores. Parallel programs should be started by using <code>mpirun</code> command. ===== Example =====Bob user allocates 2 nodes, 12 hour for an MPI job, billing 'barfoo' account. 24 MPI thread will be started on each node. The stdout output is piped to <code>slurm.out</code> file (<code>#SBATCH -o</code>). <pre>#!/bin/bash#SBATCH -A barfoo#SBATCH --job-name=mpi#SBATCH -N 2#SBATCH --ntasks-per-node=24#SBATCH --time=12:00:00#SBATCH -o slurm.outmpirun ./a.out</pre> ==== CPU binding ====Generally, the performance of MPI application can be optimized with CPU core binding. In this case, the threads of the paralel program won't be scheduled by the OS between the CPU cores, and the memory localization can be made better (less cache miss). It is advised to use memory binding. Tests can be run to define, what binding strategy gives the best performance for our application. The following settings are valid for OpenMPI environment. Further information on binding can be retrieved with <code>--report-bindings</code> MPI option. Along with the running commands, few lines of the detailed binding information are shown. It is important, that one should not use task_binding of the scheduler! ===== Binding per CPU core =====In this case, MPI fills CPU cores by the order of threads (rank). <pre>Command to run: mpirun --bind-to-core --bycore [cn05:05493] MCW rank 0 bound to socket 0[core 0]: [B . . . . . . . . . . .][. . . . . . . . . . . .][cn05:05493] MCW rank 1 bound to socket 0[core 1]: [. B . . . . . . . . . .][. . . . . . . . . . . .][cn05:05493] MCW rank 2 bound to socket 0[core 2]: [. . B . . . . . . . . .][. . . . . . . . . . . .][cn05:05493] MCW rank 3 bound to socket 0[core 3]: [. . . B . . . . . . . .][. . . . . . . . . . . .]</pre> ===== Binding based on CPU socket =====In this case, MPI threads are filling CPUs alternately.<pre>Command to run: mpirun --bind-to-core --bysocket [cn05:05659] MCW rank 0 bound to socket 0[core 0]: [B . . . . . . . . . . .][. . . . . . . . . . . .][cn05:05659] MCW rank 1 bound to socket 1[core 0]: [. . . . . . . . . . . .][B . . . . . . . . . . .][cn05:05659] MCW rank 2 bound to socket 0[core 1]: [. B . . . . . . . . . .][. . . . . . . . . . . .][cn05:05659] MCW rank 3 bound to socket 1[core 1]: [. . . . . . . . . . . .][. B . . . . . . . . . .]</pre> ===== Binding by nodes =====In this case, MPI threads are filling nodes alternately. At least 2 nodes needs to be allocated.<pre>Command to run: mpirun --bind-to-core --bynode [cn05:05904] MCW rank 0 bound to socket 0[core 0]: [B . . . . . . . . . . .][. . . . . . . . . . . .][cn05:05904] MCW rank 2 bound to socket 0[core 1]: [. B . . . . . . . . . .][. . . . . . . . . . . .][cn06:05969] MCW rank 1 bound to socket 0[core 0]: [B . . . . . . . . . . .][. . . . . . . . . . . .][cn06:05969] MCW rank 3 bound to socket 0[core 1]: [. B . . . . . . . . . .][. . . . . . . . . . . .]</pre> ==== OpenMP (OMP) jobs ====For OpenMP paralell applications, 1 node needs to be allocated, and the number of OMP threads needs to be provided with the <code>OMP_NUM_THREADS</code> environment variable. The variable needs to be written before the application (see example), or needs to be exported before executing the command:<code> export OMP_NUM_THREADS=24</code> ===== Example =====Alice user starts a 24 threaded OMP application for maximum 6 hours on the expenses of foobar account.<pre>#!/bin/bash#SBATCH -A foobar#SBATCH --job-name=omp#SBATCH --time=06:00:00#SBATCH -N 1OMP_NUM_THREADS=24 ./a.out</pre> ==== Hybrid MPI-OMP jobs ==== When an application uses MPI and OMP it is running in hybrid MPI-OMP mode. Good to know that Intel MKL linked applications MKL calls are OpenMP capable. Generally, the following distribution suggested: MPI process number is from 1 to the CPU socket number, OMP thread number is the number of CPU cores in a node, or the half or quarter of that (it depends on code). For the job script, the parameters of these two needs to be combined. ===== Example =====Alice user sent a hybrid job on the expenses of the 'foobar' account for 8 hours, and 2 nodes. 1 MPI process is running on one node using 24 OMP thread per node. For the 2 nodes, 2 MPI process is running, with 2x24 OMP threads<pre>#!/bin/bash#SBATCH -A foobar#SBATCH --job-name=mpiomp#SBATCH -N 2#SBATCH --time=08:00:00#SBATCH --ntasks-per-node=1#SBATCH -o slurm.outexport OMP_NUM_THREADS=24mpirun ./a.out</pre> ==== Maple Grid jobs ====Maple can be run - similarly to OMP jobs - on one node. Maple module need to be loaded for using it. A grid server needs to be started, because Maple is working in client-server mode (<code>${MAPLE}/toolbox/Grid/bin/startserver</code>). This application needs to use license, which have to be given in the jobscript (<code>#SBATCH --licenses=maplegrid:1</code>). Starting of a Maple job is done by using <code>${MAPLE}/toolbox/Grid/bin/joblauncher</code> code. ===== Example =====Alice user is running a Maple Grid application for 6 hours on the expenses of 'foobar' account:<pre>#!/bin/bash#SBATCH -A foobar#SBATCH --job-name=maple#SBATCH -N 1#SBATCH --ntasks-per-node=24#SBATCH --time=06:00:00#SBATCH -o slurm.out#SBATCH --licenses=maplegrid:1 module load maple ${MAPLE}/toolbox/Grid/bin/startserver${MAPLE}/toolbox/Grid/bin/joblauncher ${MAPLE}/toolbox/Grid Engine scheduler /samples/Simple.mpl</pre> ==== GPU compute nodes ====The Szeged site accomodates 2 GPU enabled compute nodes. Each GPU node has 6 Nvidia Tesla M2070 cards. The GPU nodes reside in a separate job queue (<code>--partition gpu</code>). To specify the number of GPUs set <code>--gres gpu:#</code> directive. ===== Example =====Alice user submits to the foobar account a 4 GPU, 6 hour job.<pre>#!/bin/bash#SBATCH -A foobar#SBATCH --job-name=GPU#SBATCH --partition gpu#SBATCH --gres gpu:4#SBATCH --time=06:00:00 $PWD/gpu_burnout 3600</pre> == Extensions ==Extensions should be asked for at the Execution site (NIIF) at prace-support@niif.hu. All requests will be carefully reviewed and decided if eligable.
== Reporting after finishing project ==
A report must be created after using PRACE resources. Please contact prace-support@niif.hu for further details.
PRACE
KIFÜ
Where technical support has been received the following additional text should also be used:
'''The support of [name of person/people] from KIFÜ, Hungary to the technical work is gratefully acknowledged.'''
