serpent in SGE queuing system

Parallelization with OpenMP and MPI, scalability, reproducibility, errors, problems suggestions
Post Reply
rullyh
Posts: 7
Joined: Mon Jul 16, 2012 4:04 am
Security question 1: No
Security question 2: 92

serpent in SGE queuing system

Post by rullyh » Wed Jul 17, 2013 8:04 am

Dear fellas,

We are new to the SGE queuing System

I just wondering is there any body have experience combining open mp and mpi in SGE queuing system?

We have problem while submiting serpent2 jobs in SGE system. The SGE only recognize number of core which is defined by mpi, but not by open mp.


Thanks for suggestion

orca.blu
Posts: 59
Joined: Wed Apr 20, 2011 1:39 pm

Re: serpent in SGE queuing system

Post by orca.blu » Wed Jul 17, 2013 12:15 pm

Hi rullyh,
I do not know your job scheduler, so my suggestion is probably useless.

Anyway, I had the same problem on blue gene/q: it did not recognize the omp parallelization in each node.
This is the final launching script which solved the problem.

Code: Select all

#!/bin/bash

# @ job_type = bluegene
# @ bg_connectivity = MESH
# @ wall_clock_limit = 00:30:00
# @ notification = never
# @ bg_size = 64
# @ bg_rotate = FALSE
# @ job_name = flattop_1024_64_64_16o
# @ initialdir = .
# @ account_no = IscrC_MPMCF
# @ error = $(job_name)_$(jobid).err
# @ output = $(job_name)_$(jobid).out
# @ queue

export TOTAL_MPI_PROCESSES=64
export TASK_PER_NODE=1
export THREADS=16

echo "Job started at `date`"

EXEC=/fermi/home/userexternal/maufiero/serpent/zero/serpent2.1.12_m4_xl/sss2_xl

runjob --np $TOTAL_MPI_PROCESSES --ranks-per-node $TASK_PER_NODE --env-all : $EXEC -omp $THREADS gpfs/scratch/userexternal/maufiero/input/sca_test/flattop_1024_64_64

echo "Job finished at `date`"
Manuele Aufiero
LPSC/IN2P3/CNRS Grenoble

mathieu.hursin
Posts: 34
Joined: Thu Sep 22, 2011 5:27 pm

Re: serpent in SGE queuing system

Post by mathieu.hursin » Mon Jan 27, 2014 8:57 pm

Dear all,
I m using SERPENT-v2 on a cluster where the ressources are handled by a SGE queuing system. I m running SERPENT-v2 within some scripts driving branch calculations, the end goal being the generation of few group cross sections.

Is there an easy way to obtain either from SERPENT-v2 or from the queuing system a signal telling that the SERPENT run in the queue is done?

Thanks a lot for your help,

Best regards,

Mathieu

Post Reply