Re: [AMBER] REM_hybrid test fails on AIX

From: Shan-ho Tsai <tsai.hal.physast.uga.edu>
Date: Thu, 5 Feb 2009 11:44:43 -0500 (EST)

Dear all,
I am still trying to get parallel Amber10 to build
correctly on AIX5.2, using XLF9.1. The serial build
passed all tests, but the parallel build fails
drastically the REM_hybrid test (the potential
energy already in the first replica exchange is
approx. -1450, while the correct value is approx.
-615). I get the same wrong result using the IBM
native MPI libs and using MPICH.

Given that the other REM tests passed, does it mean
that the problem is coming from having explicit and
implicit solvent in the rem_hybrid test? What portion
of the Amber10 code might be giving rise to this
error?

Any suggestions what I should test and how to get
this problem resolved?

Thank you in advance!
Shan-Ho

-----------------------
Shan-Ho Tsai
University of Georgia, Athens GA.


On Thu, 22 Jan 2009, Shan-ho Tsai wrote:

>
> Dear all,
>
> I am trying to build Amber10 using XLF 9.1 on an IBM
> power4+ node, running AIX 5.2. The serial version
> compiled fine and all tests passed. I am building the
> parallel version with
>
> ./configure_amber -mpi -nobintraj xlf90_aix
>
> (The -nobintraj option was necessary to avoid an
> error in the netcdf support. I got the hint from
> a posting in the Amber mailing-list archive.)
>
> gmake parallel
>
> There were no fatal errors, although there were
> some compilation warnings. I will send this in,
> if they are helpful in troubleshooting.
>
> I ran the tests with 2 and with 4 processors and the
> rem_hybrid test failed drastically. All other tests
> passed (for rem_gb_2rep, rem_wat, and rem_gb_4rep
> I had to replace the final "&end" by "/" in rem.in.000,
> etc, then the tests ran and passed).
>
> For rem_hybrid the differences in the results are
> shown below (sorry for the long attachment). The
> results that I get with 2 and with 4 processors
> are basically the same.
>
> Any ideas where the problem is coming from? Does this
> indicate a problem in my build of sander.MPI, or could
> it be in some other components of the package?
>
> I will appreciate any help, thank you very much!
>
> Sincerely,
> Shan-Ho Tsai
>
> --------------------
>
> cd test/rem_hybrid
> more rem.log.dif
> < 1 1.05 0. -615.76 300.00 330.00 0. -1
> < 2 0.95 0. -615.76 330.00 300.00 2.00 -1
> ---
> > 1 1.05 0. -1449.96 300.00 330.00 0. -1
> > 2 0.95 0. -1449.96 330.00 300.00 2.00 -1

.....

_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Feb 06 2009 - 01:21:45 PST
Custom Search