On Wed, Dec 14, 2011 at 2:46 PM, Reuti <re...@staff.uni-marburg.de> wrote:
> Am 14.12.2011 um 07:05 schrieb mahbube rustaee: > > > Hello all, > > > > Thanks ,excuse me for many questions and your Inconvenience. > > > > consumable resource for h_vmem: > > h_vmem h_vmem MEMORY <= FORCED YES > 0 0 > > I set h_vmem=18G on host level. > > This is what is installed? > openmpi > > > I have a program. when I run it via CLI ,it consumes about 400M per slot > (via top command). > > Maybe the application initializes a bigger memory array which it doesn't > use later on. One can mimic the behavior of SGE by setting: > > $ ulimit -H -v 409600 > No, a simple program, such hello world from any rank with 100 slots. > -- Reuti > > > > when I send it via job with: > > ... > > #$ -l h_vmem=900M > > ... > > > > job will be killed via signal 9 and output is: > > > -------------------------------------------------------------------------- > > mpirun noticed that process rank 48 with PID 20806 on node > xeon-2-5.local exited on signal 9 (Killed). > > > -------------------------------------------------------------------------- > > > > and with : > > -l h_vmem=1G (or upper) > > > > job runs successfully. I have this problem with many jobs that need a > little memory (about 20M) but I have to set h_vmem to a big value. > > > > why? > > Thx > > _______________________________________________ > > users mailing list > > users@gridengine.org > > https://gridengine.org/mailman/listinfo/users > >
_______________________________________________ users mailing list users@gridengine.org https://gridengine.org/mailman/listinfo/users
