[gmx-users] Unexplained crashes with Gromacs-4.0

Berk Hess gmx3 at hotmail.com
Tue Oct 14 13:44:24 CEST 2008





> Date: Tue, 14 Oct 2008 07:19:45 -0400
> From: jalemkul at vt.edu
> To: gmx-users at gromacs.org
> Subject: Re: [gmx-users] Unexplained crashes with Gromacs-4.0
> 
> 
> 
> Berk Hess wrote:
> > Hi,
> > 
> > I don't really know what could be the problem here.
> > 
> > 0..180 is (confusingly) not including the end, so 180 is out of range.
> > 
> > The only thing I can think of (except for unknown bugs of coarse),
> > is that I corrected the coupling time of the Parrinello-Rahman barostat
> > when going from 4.0 RC4 to the official 4.0 release.
> > I guess all your old simulations were with older Gromacs releases.
> > So to get the same coupling as before, so would need to multiply tau_t 
> > by 16.6.
> > I don't know if this would help.
> 
> I did see this same error under RC2, but noticed that you fixed this bug shortly 
> after I saw it, and that's what I assumed it was.  But now I have seen it under 
> the official 4.0 release.
> 
> > 
> > You did not get pressure coupling warnings before the crash?
> > 
> 
> There are no messages printed before this error pops up.  What's really 
> confusing is that if I simply try the same .tpr file again, it runs without a 
> problem.  I know that's not terribly scientific, but it seemed odd to me that my 
> systems running for 40+ ns suddenly had a problem.
> 
> Is there any diagnostic information I can provide?  This problem seems like it 
> would be elusive, since even I can't fully reproduce it.
> 
> -Justin

If you run in parallel, the dynamic load balancing can affective reprodibility.
You can turn it off with -dlb no

Another source of non-reproducibility is FFTW (with PME).
By default FFTW times which of different codes is the fastest and uses that.
Configuring FFTW with SSE makes all codes faster and the differences smaller,
so you can configure Gromacs with --disable-fftw-measure to remove
the run time optimization without performance loss.
For debugging purposes the new -reprod switch of mdrun will try to turn
off all optimizations which might affect reproducibilty.

Berk


_________________________________________________________________
Express yourself instantly with MSN Messenger! Download today it's FREE!
http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20081014/26c0462b/attachment.html>


More information about the gromacs.org_gmx-users mailing list