[Bioclusters] anyone using gridMathematica ?

[landman]

The documentation page
http://documents.wolfram.com/applications/parallel/Configuration/Unix.html
covers some examples where one could easily substitute a qrsh for an rsh/ssh,
or a relevant qsub.

It appears that it should be easy to "bring into the fold" given the system
they describe.  Of course, in the field, it is usually another matter.

It would appear that it would require a configuration file similar to 


LaunchSlave["remote_host","qrsh -now  -l host='l' math -mathlink"]

(syntax may not be correct for your job scheduler)

In this case, as qrsh uses its own selection mechanism, the "remote_host" may
not be relevant (though it is likely that you cannot omit it).  You should try
without the "-l host='l'" if the above does not work.

There are other "tricks" including inserting a qsub at the outset with the
required number of cpus, and a job that launches mathlink after reporting back
through some IPC mechanism, what its hostname is.  This "bubble" method is a
hack, but it does work.

Joe


On Thu, 19 Jun 2003 12:36:05 -0400, Chris Dagdigian wrote
> I'm trying to get my head around the concept of using Wolfram's 
> gridMathematica in a compute farm setting.
> 
> The Mathematica product allows one to launch "mathlink kernels" on 
> remote unix or windows hosts. They are quite proud of how standalone 
> and independent the product is (all it needs is a tcp connection). 
> heh. I'm tring to bring the product into the fold so to speak with 
> other cluster-aware applications that use the cluster resource 
> management and batch scheduling subsystems.
> 
> Single host use within a compute farm is easy. It's trivial to bsub 
> or qsub an interactive request for 'math' and run a single 
> mathematica kernel on a fast remote machine. Now I want to run 
> against N mathematica kernels speaking via mathlink to each other.
> 
> The eventual goal is tight integration within LSF and SGE but right 
> now I'll be happy with a loose "get the hostnames from the batch 
> scheduler and feed those into a mathematica notebook or shell script 
> that will then launch the mathlink kernels" sort of way.
> 
> Has anyone done this?
> 
> Also- I'm a complete idiot when it comes to math. Are there any 
> mathematica users out there who can share a test/demo computation 
> that actually makes use of functions like ParallelMap and ParallelTable?
> 
> Regards,
> Chris
> 
> _______________________________________________
> Bioclusters maillist  -  Bioclusters@bioinformatics.org
> https://bioinformatics.org/mailman/listinfo/bioclusters


--
Joseph Landman, Ph.D
Scalable Informatics LLC,
email: landman@scalableinformatics.com
web  : http://scalableinformatics.com
phone: +1 734 612 4615