[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [condor-users] Condor-G question



> 
> >        If i want to run N copies of the executable but with different
> >arguments (like a parameter sweep) but still wanting to fork only one
> >jobmanager process for those N copies how will i do it. Any ideas, hidden
> >condor features that we could use.
> 
> If the remote jobmanager could be anything: Condor, PBS, LSF, or whatever, 
> then I don't think there is a way to currently do this.
> 
> If you know that the remote jobmanager is Condor, there is no automatic way 
> to do it. (Anyone want to take bets on Jaime Frey posting some cleve method 
> for this just to show me up?) But, you could submit a job that is a script 
> that:
> 
>   1) Submits N jobs to Condor
>   2) Collates results from these jobs.
>   3) Exits only after these jobs finish.
> 
> I suspect that this would be prone to difficulties occurred: What is the 
> script fails but the jobs don't? What if Condor-G resubmits your job, and 
> the previous jobs are still running?

Instead of a script why not make it a DAGman job that spawns the other
jobs, and then cleans up when done?

Scott

> 
> Also, this is Condor specific, and that seems inelegant if you want to work 
> in a heterogeneous grid.
> 
> -alain
> 
> 
> Condor Support Information:
> http://www.cs.wisc.edu/condor/condor-support/
> To Unsubscribe, send mail to majordomo@xxxxxxxxxxx with
> unsubscribe condor-users <your_email_address>
Condor Support Information:
http://www.cs.wisc.edu/condor/condor-support/
To Unsubscribe, send mail to majordomo@xxxxxxxxxxx with
unsubscribe condor-users <your_email_address>