AMBER Archive (2008)

Subject: Re: AMBER: sander.MPI setup with SGE

From: Sasha Buzko (obuzko_at_ucla.edu)
Date: Mon Apr 07 2008 - 15:51:20 CDT


Hi Ross,
thanks for your suggestions.

I did change the number of slots from 20 to 16, and the nodes do have
the same environment as the compilation unit (in fact, I compiled Amber
on one of the nodes to keep environment issues minimal). In any event,
judging from some testing, it looks like the issue is in the SGE
scheduling that handles MPI environment. I'm trying to get some help
from the IT people at UCLA, and will let you know if any Amber-specific
issues arise once the SGE setup is fixed.

Thanks

Sasha

Ross Walker wrote:
> Hi Sasha,
>
> It's good to hear from you.
>
> It's a little hard to help out much with the problems you are seeing since
> there isn't really enough information. A couple of points though. Firstly
> with sander.MPI always try to use a power of 2 cpus, so in this case 16
> rather than 20. This will make a big difference on performance. With PMEMD
> it is not as critical.
>
> My second guess to the problem would be one of the environment not being
> identical on the compute nodes to the compilation node such that there are
> missing dynamic libraries needed by sander.MPI. You could try compiling
> things statically and see if this helps - not always possible due to
> restrictions on mpi libraries etc but worth a try. Add
>
> ./configure -static ....
>
> Then recompile.
>
> I'd also try something simple as well to make sure it is sander causing the
> problem. I.e. try just running 'ls' instead of sander and see if you get
> output from twenty copies of ls.
>
> As for error messages there should be a job submission output file somewhere
> that any error message will likely have gone into. With pbs this starts in a
> directory called ~/.pbs_spool and then ultimately gets copied into the
> directory from which the job submission was done. I'm not sure where sun
> grid engine puts things but I would search around for that.
>
> I'm sorry I can't help much more than that but if you can track down where
> the error message went it will be much easier to help.
>
> All the best
> Ross
>
> /\
> \/
> |\oss Walker
>
> | Assistant Research Professor |
> | San Diego Supercomputer Center |
> | Tel: +1 858 822 0854 | EMail:- ross_at_rosswalker.co.uk |
> | http://www.rosswalker.co.uk | PGP Key available on request |
>
> Note: Electronic Mail is not secure, has no guarantee of delivery, may not
> be read every day, and should not be used for urgent or sensitive issues.
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu