Re: [AMBER] Restarting PMEMD calculation after server shutdown

From: Jason Swails <>
Date: Thu, 2 Aug 2012 08:17:38 -0400

On Thu, Aug 2, 2012 at 8:05 AM, David A Case <>wrote:

> On Thu, Aug 02, 2012, Saaz Sakrikar wrote:
> > I have faced a recurring problem when trying to do a pmemd simulation on
> my
> > server. It is a long calculation (nstlim = 15 million)
> >
> > The server has many shutdowns and thrice the calculation has been stopped
> > in the middle of the run. Is it possible to resart it from where it left
> > off and not from scratch? Or is that only for Sander calculations?
> > I have saved mdcrd and rst files from these attempts.
> You can start with any rst file that was correctly written, but it is
> likely
> to be a bit (or a lot) of a pain to combine all the mdcrd files if jobs
> were
> stopped in the middle.
> I rarely (=never) try to run really long calculations in a single run, but
> rather plan in advance to split things into bite-size pieces, no more than
> 24-48 hours each. That way, if the server fails, you lose at most 1-2
> days.
> Depending on how you access the server, you may be able to create a simple
> script that will automatically submit a second job when the first one is
> completed.

As Dave mentions above, it is typically much better to run 15 jobs with 1
million steps each (each starting where the last left off) rather than run
1 job with 15 million steps.

If the cluster you are running on uses torque or some other PBS-derived
queuing system, there's an easy mechanism where you can just submit all 15
jobs, forget about them, and have them all run consecutively.

I put together a little how-to use PBS awhile back which may be helpful
(specifically the mechanism where you submit all jobs at once and have them
run in a specified order). You can access it here: (with the mechanism I mentioned
described here:


Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
AMBER mailing list
Received on Thu Aug 02 2012 - 05:30:05 PDT
Custom Search