Re: [AMBER] Problems when I wanted to restart the MD production

From: Jason Swails <jason.swails.gmail.com>
Date: Thu, 4 Aug 2011 10:44:32 -0400

On Aug 4, 2011, at 1:48 AM, Bill Ross <ross.cgl.ucsf.EDU> wrote:

> Back in single-cpu days, mdinfo was flushed immediately, is what
> I remember, while mdout was flushed seldom, if at all. Not sure
> how relevant that is now, but the idea was that the cost of
> synchronizing with disk would be limited to a quick write.

That may be true for sander, but sander still uses the unreliable flush intrinsic that doesn't work on kraken (and I'm guessing other supercomputers). pmemd does a hard flush by closing the file and reopening it, and has 2 variables: mdout_flush_interval and mdinfo_flush_interval. The first is 30 by default and the second is 60.

Therefore, pmemd is the only program that I could reliably predict this condition with.

All the best,
Jason

>
> Bill
>
>> that is, output is flushed to the mdout file at least
>> every 30 seconds, whereas the mdinfo is flushed every 60 seconds). This is
>> done for performance reasons, but also shows that the mdinfo file may show
>> different information than the mdout file.
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber

_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Aug 04 2011 - 08:00:03 PDT
Custom Search