Re: [AMBER] Odd coordinate writing problem in pmemd.MPI

From: <>
Date: Thu, 5 Apr 2012 16:33:05 -0400 (EDT)

Hi Ross,

> Is this reproducible? Or does it appear to be random?

I'll check about that. I never experienced this problem before but this
may be the first time that I write such a large trajectory quickly and in
one piece. Other people certainly must have though and this issue would be
very noticeable.

> Is there anything magical about the point at which it happens as well.
> if it occurs almost exactly when the trajectory reaches 2GB in size (or
> 4GB)
> then it is almost certainly some corruption due to use of 32 bit ints in

good point, that looks possible. The crd-file is about 10GB large and
scrambled coords start at 41% of the file onwards.

> file handling. It would take some searching to locate this, it could be
> anywhere from within the pmemd code, in the MPI library or in the
> operating
> system I/O library (maybe even the compiler used). Can you confirm that
> you
> are running a 64 bit OS with 64 bit compilers, and 64 bit MPI?

yes, yes (gcc from Fedora x86_64) and I believe yes. Openmpi state that
they support 32 as well as 64 bit OS and the library was compiled from

> Also are you writing locally or to a remote file system?

This is locally on the same machine.

I will see if I can find a way to reliably reproduce the problem and
isolate where it is coming from (amber, compilers or mpi libraries...).


Dr. Thomas Steinbrecher
formerly at the
BioMaps Institute
Rutgers University
610 Taylor Rd.
Piscataway, NJ 08854

AMBER mailing list
Received on Thu Apr 05 2012 - 14:00:04 PDT
Custom Search