I have fixed the issue on our cluster by reverting previous config changes I made to support large task arrays: max_aj_instances = 0 max_aj_tasks = 500000
Reverting these back to their defaults of 2000 and 75000 allowed -tc to work properly. I was also able to raise max_aj_tasks back to 500000. According to the docs, setting max_aj_instances to zero deactivates the limit of max array tasks to run simultaneously per array job. It may be that it also bypasses the user-specified -tc limit. Best, Chris PS Thanks to Hugh MacMullan @upenn for privately pointing out -tc worked fine in a default cluster which led me to hunting down changes. On Mon, Sep 16, 2013 at 12:54 PM, Chris Black <[email protected]> wrote: > Greetings! We are trying to move from SGE 6.2u4 to Son of GridEngine 8.1.x > and all is going fairly well except for qsub -tc. > > The max running task limit option to qsub does not seem to be honored on > Son of GridEngine 8.1.x. > > I have a simple sample job and can confirm the same same qsub leads to > proper limiting of concurrent task array tasks on our older Sun SGE 6.2u4 > cluster. > > The qsub cmdline I am using is: > > qsub -q rnd.q -t 1-50 -tc 2 -N cbTA -j y -cwd /path/to/task.sh > > task.sh just contains some echos and a sleep 600. > > On our 6.2u4 cluster, the scheduler properly only runs two of the 50 tasks > at once. On SoGE 8.1.x all run at once if the resources are available. > Clusters are separate and do not share nodes. > > We were running SoGE 8.1.3 and we updated to 8.1.4 last week and the issue > is still there. > > Is qsub -tc broken in SoGE 8.1.4 or am I missing some scheduler config > option I need to change to enable it? Any other thoughts or next steps for > solving this? > > Thanks in advance, > Chris > > -- > Chris Black > Linux Administrator > Sequencing Solutions > -- Chris Black Linux Administrator Sequencing Solutions
_______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
