Hi Anselm,
we observed the same thing recently and found that the explanaitions
from the following mail in the amber mailing list archive covers about
all your questions:
http://amber.ch.ic.ac.uk/archive/200401/0103.html
Benjamin
Anselm Horn schrieb:
> Dear all,
>
> we encountered a strange behaviour during a pmemd restart on our cluster
> (infiniband network, job ran on 8 nodes with 4 processors each):
>
> A restart using a certain rst file crashed because there were numbers
> too large for the output format and thus only asterics appeared (known
> problem). When we reran the MD calculation starting from the previous
> rst file that also served as input for the trajectory that produced the
> corrupted rst file, pmemd then produced a valid rst file at the end of
> this new trajectory. Thus, the same rst file on the same (homogeneous)
> cluster with the same executable yielded a different result. How can
> this be explained?
>
> Is it possible that the parallel implementation is the reason for this
> non-deterministic behaviour - or does this simply indicate a potential
> hardware problem on some of the machines?
>
> Any hints are welcome.
>
> Regards,
>
> Anselm Horn
> Bioinformatik
> Emil-Fischer-Zentrum
> Friedrich-Alexander-Universität Erlangen-Nürnberg
>
>
>
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber.scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
>
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Wed Dec 19 2007 - 06:07:15 PST
