AMBER Archive (2007)Subject: Re: AMBER: OpenMPI on MacBook Pro -- problems
From: Mike Summers (summers_at_hhmi.umbc.edu)
Date: Fri Mar 02 2007 - 14:04:17 CST
David,
I've been using web demos as a guide while I learn to use Amber.
I guess my procedure still needs tweaking. Thanks for the advice.
Regarding the MacBook Pro failures:
1. The SAME job runs fine on the MacBook Pro in Serial mode.
2. The SAME job runs fine on our Linux cluster in Parallel mode.
3. A different job (minimization) runs fine on the MacBook Pro in Parallel mode.
5. The make test.serial runs to completion with some small differences (errors).
4. The make test.parallel does NOT run to completion. Some of the output is shown
below:
################################################################
==============================================================
diffing poh.out.save with poh.out
PASSED
==============================================================
diffing evbout.save with evbout
PASSED
==============================================================
cd rem_gb_2rep; ./Run.rem
Running multisander version of sander amber9
Total processors = 2
Number of groups = 2
Looping over processors:
WorldRank is the global PE rank
NodeID is the local PE rank in current group
Group = 0
WorldRank = 0
NodeID = 0
Group = 1
WorldRank = 1
NodeID = 0
mpirun noticed that job rank 0 with PID 26642 on node "localhost" exited on signal 64.
1 additional process aborted (not shown)
diffing rem.log.save with rem.log
possible FAILURE: check rem.log.dif
==============================================================
diffing reminfo.000.save with reminfo.000
possible FAILURE: check reminfo.000.dif
==============================================================
cd rem_gb_4rep; ./Run.rem
This test requires at least 4 threads...skipping test
cd rem_wat; ./Run.rem
csh: /bin/rm: No match.
#############################################################################
Thanks,
Mike
On Fri, Mar 02, 2007 at 09:36:28AM -0800, David A. Case wrote:
> On Fri, Mar 02, 2007, Mike Summers wrote:
>
> > I'm having a strange problem using Amber9 with OpenMPI-1.1.4 on a MacBook Pro
>
> > I have no problem running in MPI mode on our linux cluster, which has only 1G
> > memory per node (compared to 2 G on the MacBook).
>
> Mike:
>
> Just to be sure: are you saying that the *same* job fails on the Mac but runs
> correctly on the Linux cluster? And does this job work OK in serial but not
> in parallel?
>
> Does the Mac version pass the test suite in parallel?
>
> How many atoms are in your system?
>
> [On a more general note, your system looks pretty dangerous (and may be
> related to the error message you got). You are running at
> constant pressure, but starting with irest=0. This is dangerous unless you
> actually have equilibrated coordinates -- the general rule is to equilibrate
> at constant volume, then use a restart to switch to constant pressure. Also,
> the job plans to cool the system down to 0K at the end; this probably won't
> work with constant pressure, since the calculation of pressure is very
> unstable at low temperature. For this and other reasons, simulated annealing
> runs that cool down to 0K generally don't work well for explicit solvent
> simulations.
>
> Another wild guess: You file names are things like "capwat...". Do you want
> to just solvate part of the system? If so, set ntb=0 and use the fcap
> parameter.]
>
>
> ...good luck...dave
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
--
*********************************
Michael F. Summers
Department of Chemistry and Biochemistry
and Howard Hughes Medical Institute
University of Maryland Baltimore County
1000 Hilltop Circle
Baltimore, MD 21250
Phone: (410)-455-2527
FAX: (410)-455-1174
Email: summers_at_hhmi.umbc.edu
Web: www.hhmi.umbc.edu
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
|