Re: [AMBER-Developers] infinite ptraj.MPI, was: First AmberTools release candidate

From: case <case.biomaps.rutgers.edu>
Date: Wed, 17 Mar 2010 07:38:44 -0400

On Tue, Mar 16, 2010, Ross Walker wrote:
>
> If you are trying to run MPI/IO over NFS then I'm not surprised it runs
> slower. NFS is utter junk for anything other than serial reads and write.
> All you are doing is thrashing things. Probably even more so if the server
> on the other end is not a nice spec raid system which can actually do
> parallel read and writes properly without adding in 10ms of disk head seek
> to every call. You really need a local (sata / SAS based) raid system or a
> real parallel filesystem like luster or GPFS to see any benefit of running
> ptraj, and thus the I/O in parallel.

Dan: can you volunteer to craft a paragraph along the lines of the above, plus
your experience, to put into the Users' Manual. We want to warn people to be
careful about using parallel ptraj, and certainly to check its performance vs.
serial, etc.

..thx...dave


_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Wed Mar 17 2010 - 05:00:02 PDT
Custom Search