[GE users] can't start pthread with Gridengine

echoecho chrisbee at u.washington.edu
Mon Jul 27 20:50:45 BST 2009


Looking at resource requirements is good idea.  I have vmem set as a consumable resource on each node.  Every job submits with a default value for h_vmem.  I'll try disabling this for one node and see what happens.


> Hi Chris,
> 
> its just a wild guess, but we ran into trouble with Matlab and other 
> applications with older versions of SGE when submitting jobs with a h_rt 
> limit (which in the older versions also set the stack size to the size 
> specified for h_rt -> see "ulimit -a"). I heard somewhere that this 
> setting was also causing pthreaded application to crash, though I never 
> verified that myself.
> 
> As I said, just a wild guess. But would you let me know, if this was 
> indeed the problem?
> Best,
> Sabine
> 
> echoecho schrieb:
> > Yes something is different about the environment when jobs are started by SGE.  I saved the output from env generated by a bash script submitted through qsub, then ssh'ed into a node directly, loaded the saved environemnt variables and tried my python pthreads test script.  It successfully created a thread so I don't think any difference in environment variables is the cause.
> >
> > Are there any special configurations that need to be made to get Gridengine to play nice with libpthreads?
> >
> > Chris
> >
> >   
> >> SGE doesn't do anything to your scripts other than execute them.  
> >> Perhaps there's something different about the shell or environment that 
> >> is being started on the exec node.  Try running "qrsh env" to see how 
> >> the environment variables are set when you're job is executing on the 
> >> remote machine.
> >>
> >> Daniel
> >>
> >> echoecho wrote:
> >>     
> >>> Hello,
> >>>
> >>> I'd like to run multi-threaded applications on our cluster as Gridengine jobs, but it appears that Gridengine is not allowing the creation of pthreads.  I've tested two applications: BWA (short-read DNA sequence alignment) and a simple python script that makes one thread.  In both cases the programs cannot create pthreads.  I've tried with and without a parallel environment set up with this configuration:
> >>>
> >>> $ qconf -sp smp
> >>> pe_name           smp
> >>> slots             999
> >>> user_lists        NONE
> >>> xuser_lists       NONE
> >>> start_proc_args   /bin/true
> >>> stop_proc_args    /bin/true
> >>> allocation_rule   $pe_slots
> >>> control_slaves    FALSE
> >>> job_is_first_task TRUE
> >>> urgency_slots     min
> >>>
> >>> Both applications can create threads if started directly on a compute node outside of SGE.
> >>>
> >>> Attached is the python script I've used to test thread creation and the bash script that creates the submit script and calls qsub.
> >>>
> >>> Any help in troubleshooting this problem would be much appreciated.
> >>>
> >>> Thanks,
> >>> Chris
> >>>
> >>> --
> >>> Chris Berthiaume
> >>> University of Washington
> >>> Center for Environmental Genomics
> >>>
> >>> ------------------------------------------------------
> >>> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=209228
> >>>
> >>> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
> >>>
> >>>       
> >
> > ------------------------------------------------------
> > http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=209269
> >
> > To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
> >
> >

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=209773

To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].



More information about the gridengine-users mailing list