Re: AMBER: Middlebury College Cluster and Amber 8

From: Robert Duke <>
Date: Fri, 11 Aug 2006 13:33:13 -0400

Peter -
Depending on how big the simulated system is, whether you are running
generalized Born or PME, etc. there are upper limits on the number of
processors allowed that relate to how the workload must be divided. I don't
know which test you are running, but a number of the tests are relatively
small systems, and parallel tests on these system are probably pretty
commonly run on 2 or 4 processors. So use a smaller processor count. If
you want to see how your system does with a larger number of processors, try
the jac and factor_ix benchmarks. These are medium to big explicit solvent
particle mesh ewald simulations, and you can use 80-128 processors without
hitting any residue limits I am sure. If you want better performance at
higher processor count, compile pmemd.
Regards - Bob Duke

----- Original Message -----
From: "Lubans, Peter S" <>
To: <>
Sent: Friday, August 11, 2006 12:55 PM
Subject: AMBER: Middlebury College Cluster and Amber 8

> Dear Amber Support,
> I am attempting to install and test Amber 8 on a ten-node dual-processor
> dual-core opteron cluster running MPICH2 1.0.4 compiled with the PGI
> compiler suite. I have tested this MPICH2 build and it passes all of its
> tests except for two that are related to our automounter.
> When I run mpdboot and specify that mpd should use either two processors
> on the local machine or one processor on each of two nodes (and set
> DO_PARALLEL="/opt/mpich2/mpich2-pgi-64/bin/mpiexec -n 2"), the parallel
> testing passes. However, if I tell the mpd system to use more than two
> processors (four on the headnode, two processors on each of two nodes,
> four processors on each of two nodes, etc) and set DO_PARALLEL to use the
> same number of processors, then one of the sander tests quits by calling
> MPI_Abort and complains that there must be more residues than processors.
> I'm not a chemist; I work in the IT department. What does this error
> message mean?
> I tried to alleviate the situation by setting DO_PARALLEL to run more
> processes than the number of processors specified when I run mpdboot.
> When I do this, Amber's tests appear to hang right at the beginning.
> Sometimes it gets through the first two, on other runs it hangs
> immediately.
> Does anyone have any ideas or suggestions?
> Sincerely,
> Peter Lubans
> Department of Library and Information Services.
> Middlebury College
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to
> To unsubscribe, send "unsubscribe amber" to

The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to
Received on Sun Aug 13 2006 - 06:07:14 PDT
Custom Search