[GE users] SGE capability question

Daniel Templeton Dan.Templeton at Sun.COM
Fri Dec 14 16:24:42 GMT 2007


    [ The following text is in the "ISO-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

And the missing piece to the serial answer that Reuti failed to mention 
is the -sync option for qsub.  If you run qsub -sync y job.sh, the qsub 
command won't return until the job completes.

Daniel

Reuti wrote:
> Hi,
>
> Am 14.12.2007 um 13:51 schrieb Dan McMahill:
>
>> I've been reading various docs and am a bit overwhelmed.  Before I go 
>> further, I have a basic question about capability.
>>
>> We have SGE deployed and I use qsub to submit jobs all the time.  
>> qstat to check on jobs, and qdel to remove them from the queue.  So 
>> far the jobs have been fairly self contained.  But now I'm interested 
>> in writing a short program probably in perl (but ruby or maybe just 
>> /bin/sh would be ok) that has a way of submitting jobs and monitoring 
>> their status. Is this something SGE can easily do?  If so, what 
>> documents should I be reading to get going?
>>
>> If I were just writing my script for a multi-processor machine, I'd 
>> just use fork/join to keep N jobs running at a time and monitor their 
>> status with the
>
> a) parallel: this could be similar while using SGE: you have to create 
> a PE and spread your job's tasks to serveral nodes (given by the 
> hostlist of granted slots) by fork in a programm and (q)rsh to another 
> node.
>
>> parent process, but I'm not sure how to do that with grid engine.  In 
>> other words, I can make my program submit all of these jobs to the 
>> queue, but I don't know how to automatically monitor the results of 
>> the job short of some hacks where I have a file in a shared file 
>> system that I monitor.
>>
>> The application is running the same calculation (which is done via an 
>> external program that may run for hours at a time) for different sets 
>> of input parameters and after all jobs have completed, assembling the 
>> results.  Actually it would be neat if results were assembled as each 
>> piece completed so I could get a partial picture along the way.
>
> b) serial: if all is serial, you could just use a for/while-loop in 
> the jobscript.
>
> -- Reuti
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net
>

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list