Re: LSF jobs

From: Djamel Ouerdane (ouerdane@nbi.dk)
Date: Wed Mar 26 2003 - 10:30:14 EST

  • Next message: Claus O. E. Jorgensen: "RICH in brag" 
    > As far as I can tell, Ionut submitted the jobs with the option 
> `-R ncpus' which will tell LSF to start the particular job on many
> CPUs.  However, Ionut did not specify how many CPUs, so it defaults to
> some number, which I guess is 1-20.  You can tell that from the output
> of `bsub -l <job#>'.  
> 
> Ionut, I'm not so sure that starting the job on several CPUs make
> sense.  I looked into your files `/brahms/u/aic/work/work3/run.sh' and 
> `/brahms/u/aic/work/work3/calibrate.cpp' and it doesn't seem to be a
> parallel program.  Hence, what will happen is, that the same (exact
> same) job will be executed on 1-20 hosts, each overwriting the same
> output file `calibData_run00<xxxx>_seq<x>.root'.  For the use of the
> `ncpus' resource to make sense, I think you have to use some sort of
> parallel execution environment, like PROOF or MLP, and your code must
> be written to use that environment.  Your `calibrate.cpp' does not
> look like that. 
> 
> Perhaps a few words on what you're trying to do would help us out.
> 


Meanwhile, some important jobs are pending. How can we wind around to 
avoid this situation ?

Djam

    This archive was generated by hypermail 2.1.5 : Wed Mar 26 2003 - 10:30:57 EST