Mpich2-intel

Download

[email protected]

September 2010

3 participants
2 discussions

Slurm integration into Hydra
by Kiryanov, Yury 28 Sep '10

28 Sep '10

Hello everybody, We noticed an issue related processes allocation by using -bootstrap slurm. I am not an expert of Slurm, but it looks like SLURM_JOB_CPUS_PER_NODE has static value and based on <slurm_installdir>/etc/slurm.conf (see Procs compute nodes attribute into it). I guess that SLURM_TASKS_PER_NODE environment variable should be handled instead of the previous one. See examples below. Example 1: $ cat job #!/bin/sh echo SLURM_TASKS_PER_NODE = $SLURM_TASKS_PER_NODE echo SLURM_JOB_CPUS_PER_NODE = $SLURM_JOB_CPUS_PER_NODE mpiexec.hydra -bootstrap slurm -n 2 hostname $ sbatch -N 2 job Submitted batch job 150 $ cat slurm-150.out SLURM_TASKS_PER_NODE = 2(x2) SLURM_JOB_CPUS_PER_NODE = 2(x2) svsmpicl104 svsmpicl104 Example2: $ sbatch -N 2 -n 2 job Submitted batch job 151 $ cat slurm-151.out SLURM_TASKS_PER_NODE = 1(x2) SLURM_JOB_CPUS_PER_NODE = 2(x2) svsmpicl104 svsmpicl104 But in the second example there were 1 processor allocated to the job on each of two nodes. It works correct if I used -ppn <ppn> into job mpiexec.hydra command line. -- Best regards, Yury Kiryanov phone: 8-83130-70700 (ext. 2531) i-net : 8313-2531 e-mail: yury.kiryanov(a)intel.com<mailto:[email protected]> -------------------------------------------------------------------- Closed Joint Stock Company Intel A/O Registered legal address: Krylatsky Hills Business Park, 17 Krylatskaya Str., Bldg 4, Moscow 121614, Russian Federation This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.

2 10

Support for Asynchronous Communication Progress in MPICH2 1.3
by Truschin, Vladimir 07 Sep '10

07 Sep '10

Hi all, Could somebody to answer on questions regarding the subject, please? 1. Do you plan implementation of Asynchronous Communication Progress for Single Threaded library? 2. As I understood, the thread of Asynchronous Communication Progress will be spin on core for Multi Threaded library. So, either separate core is needed for this thread or this thread will be negatively to impact on performance. Do you plan to improve such implementation? 3. When final release MPICH2 1.3 is expected? Best regards, Vladimir. -------------------------------------------------------------------- Closed Joint Stock Company Intel A/O Registered legal address: Krylatsky Hills Business Park, 17 Krylatskaya Str., Bldg 4, Moscow 121614, Russian Federation This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.

2 2