It probably won't help sander much; this is way down in the list of
performance-eating hogs in sander; I don't have the numbers in front of me,
but as a wild guess, if you have decent hardware and are up in the range of
128-256 processors on pmemd, you might see a 5% impact from writing to disk
too much; you could probably push it up to 10% if you tried (but realize, if
you are using 100 processors then, you are wasting 10 of them). Unless
sander has changed a lot recently, I think it is a waste of time to run over
64 procs, if that many.
Regards - Bob
----- Original Message -----
From: "Jason Swails" <jason.swails.gmail.com>
To: "AMBER Developers Mailing List" <AMBER-Developers.ambermd.org>
Sent: Tuesday, December 15, 2009 12:22 PM
Subject: [AMBER-Developers] Re: [AMBER] Number of Cycles
Hello,
This is a fork of a discussion in the regular amber mailing list, just
asking some quick opinions.
On Tue, Dec 15, 2009 at 12:09 PM, Robert Duke <rduke.email.unc.edu> wrote:
> Hmmm, did I misspeak somewhere here? Changing processor count changes
> ntwr,
> which controls the restart file write frequency; it does not touch ntwx,
> which is the frequency of dumping trajectory coordinates. Completely
> different deal. I have to leave ntwx pretty much alone, or it would have
> the potential to really muck things up for folks. The restart is of course
> for restarts, and is dumped periodically to prevent total loss of a run in
> the event of some failure, but auto-setting the frequency of writes to
> this
> file is purely a tradeoff between performance and paranoia about losing
> data
> in a crash. So as an aside, it is true that too frequently writing the
> trajectory file also has potential to whack performance, but there are
> limits to what you can do about it, and it is cheaper to write than the
> restart (no velocities, less precision). The default for restart rewrites
> in sander is probably excessively cautious.
> Regards - Bob
Scaling ntwr is a 3-line addition to mdread.f ... should we do it? Or
is only pmemd fast enough to notice the difference? :).
Thanks!
Jason
--
---------------------------------------
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Dec 15 2009 - 10:00:04 PST
