Advanced MPI scheduling: Difference between revisions

From Alliance Doc
Jump to navigation Jump to search
No edit summary
Line 1: Line 1:
{{Draft}}
{{Draft}}


Most users should submit MPI or distributed memory parallel jobs as illustrated
Most users should submit MPI or distributed memory parallel jobs following the example
at [[Running_jobs#MPI_job|Running jobs: MPI job]]. Simply request a number of  
given at [[Running_jobs#MPI_job|Running jobs]]. Simply request a number of  
processes with <code>--ntasks</code> or <code>-n</code> and trust the scheduler  
processes with <code>--ntasks</code> or <code>-n</code> and trust the scheduler  
to allocate those processes in a way that balances the efficiency of your job  
to allocate those processes in a way that balances the efficiency of your job  
with the overall efficiency of the cluster.
with the overall efficiency of the cluster.


If you need more detailed control over how your job is allocated, then read on
If you want more control over how your job is allocated, then SchedMD's
to learn about SLURM's [https://slurm.schedmd.com/sbatch.html <code>sbatch</code>]
page on [https://slurm.schedmd.com/mc_support.html multicore support] is a good
command and how its numerous options constrain the placement of processes.
place to begin. It describes how many of the options to the
[https://slurm.schedmd.com/sbatch.html <code>sbatch</code>]
command interact to constrain the placement of processes.


=== Controlling the distribution of processes ===
You may find this discussion of [https://slurm.schedmd.com/faq.html#cpu_count What exactly is considered a CPU?] in SLURM to be useful.
The basic MPI job shown at [[Running jobs]] is suitable for the majority of users, those who simply want their calculations to start running at the earliest opportunity.
 
However, you may wish to investigate how the performance of your MPI application is affected when the processes are distributed in different ways, or you may know from such an investigation that it performs best under certain constraints. For especially demanding applications this can extend to the level of not just nodes, but sockets, cores, and threads. See SchedMD's page on [https://slurm.schedmd.com/mc_support.html multicore support] for detailed information about process placement with SLURM.


=== Hybrid jobs: MPI and OpenMP, or MPI and threads ===
=== Hybrid jobs: MPI and OpenMP, or MPI and threads ===
To come
To come
You may find this discussion of [https://slurm.schedmd.com/faq.html#cpu_count What exactly is considered a CPU?] in SLURM to be useful.


=== MPI and GPUs ===
=== MPI and GPUs ===

Revision as of 18:45, 5 April 2017


This article is a draft

This is not a complete article: This is a draft, a work in progress that is intended to be published into an article, which may or may not be ready for inclusion in the main wiki. It should not necessarily be considered factual or authoritative.




Most users should submit MPI or distributed memory parallel jobs following the example given at Running jobs. Simply request a number of processes with --ntasks or -n and trust the scheduler to allocate those processes in a way that balances the efficiency of your job with the overall efficiency of the cluster.

If you want more control over how your job is allocated, then SchedMD's page on multicore support is a good place to begin. It describes how many of the options to the sbatch command interact to constrain the placement of processes.

You may find this discussion of What exactly is considered a CPU? in SLURM to be useful.

Hybrid jobs: MPI and OpenMP, or MPI and threads[edit]

To come

MPI and GPUs[edit]

To come

Troubleshooting and performance monitoring[edit]

To come

Why srun instead of mpiexec or mpirun?[edit]

To come

External links[edit]

Message Passing Interface

Retrieved from "https://docs.alliancecan.ca/mediawiki/index.php?title=Advanced_MPI_scheduling&oldid=11107"