Re: TMD: different behaviour between serial and mpi versions

From: Thomas Gaillard (
Date: Tue Jun 05 2007 - 19:25:48 CDT

On Tue, Jun 05, 2007 at 12:27:40PM -0400, Jerome Henin wrote :
> Thomas,
> Thank you for reporting this, it confirms what our tests indicated. We have
> reported it to Jim, and indeed he seemed to have found some problem with the
> code.

I have made some additional tests. I have run the TMD simulation with
the same MPI namd 2.6 binary, but with a different number of processors.

And the result (current RMSD evolution) is dependent on the number of
processors! Whereas using 1 processor with the MPI binary give the nice
linear evolution of the RMSD, 2 and 4 processors give the bad evolution
(current RMSD can't follow target RMSD), and using 2 or 4 processors
lead to a different evolution...

By the way, increasing the number of processors from 2 to 4 seems to
slightly reduce the problem.

Something interesting is that the temperature evolution is also bad
(don't stay at 300K but increase) with multiple processors. It might be
the cause of the TMD problem.

Could this be linked to a parameter dependent on the number of
processors in the code ?

> One thing I should check: in your simulations, do you use any other type of
> scripted forces, restraints etc?

I use 'rigidigidBonds all', but I think the problem also occured without
this parameter.

Thanks for your help,

Thomas Gaillard
Laboratoire de Biophysicochimie Moléculaire
Institut de Chimie LC3 - UMR 7177
Université Louis Pasteur
4 rue Blaise Pascal
F-67000 Strasbourg
phone: +33 (0)390 241 493
fax: +33 (0)390 241 490

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:44:48 CST