RE: [AMBER] AMBER is running silently

From: Ross Walker <ross.rosswalker.co.uk>
Date: Tue, 18 Aug 2009 00:33:09 +0100

Hi Kefa,

At what point does it cease producing output? Do you get anything at all or
just a completely blank mdout file?

Also if you kill the job in the queue do you ultimately get any mdout file,
I.e. was it just buffering it. What about the queue system Job output and
error files? Do these have anything?

A few things to try.

1) If you use sander vs pmemd is the behavior the same? You should probably
use PMEMD here, you will get much better performance.

2) Hot starting a membrane system at 310k is VERY dicey!. You should
probably do some careful heating at constant volume first.

3) Set nstlim=100 and ntwx=100 and ntpr=100 and run it - do you get any
output?

4) You are running periodic boundaries are you certain your inpcrd file has
box information in it? Sander and PMEMD can hang if they expect box info but
do not find it.

5) Have you run this in serial on your desktop - just to see if you get any
output - does it work here?

6) Are you certain you aren't just running into a bad node? Clusters can be
notoriously unreliable and often queuing systems do not help by keep
assigning you the same batch of nodes containing the bad node. You can
eliminate this possibility by confirming you can run in serial and/or
interactively to see if you get output.

7) Cut of 12 is pretty large. You can probably reduce this to 9.0 or even
maybe 8.0 and save yourself 50% or more in computation cost.

8) Are you using too many processors? Performance in parallel only increases
to a point (hardware, problem, compiler blah blah blah specific) and then it
falls off massively. Thus if you ask for too many cpus you can end up
running much much slower than even serial. You should benchmark you run, say
1000 steps or so on different cpu counts to see how the performance varies
and choose the optimal performance vs computational cost tradeoff that you
are prepared to stomach.

All the best
Ross

> -----Original Message-----
> From: amber-bounces.ambermd.org [mailto:amber-bounces.ambermd.org] On
> Behalf Of Kefa Lu
> Sent: Monday, August 17, 2009 4:10 PM
> To: AMBER Mailing List
> Subject: Re: [AMBER] AMBER is running silently
>
> My .mdin file is simply as following,
>
> &cntrl
> imin=0,irest=0,ntx=1,
> cut=12.0,scee=1.2,
> tempi=310.0,temp0=310.0,
> ntt=3,gamma_ln=1.0,
> ntp=2,pres0=1.0,taup=5.0,
> vlimit=25.0,
> ntb=2,ntc=2,ntf=2,
> nstlim=1000000,
> iwrap=1,
> ntwe=100,ntwx=1000,ntpr=100,
> &end
>
> Really appreciate for any suggestions.
>
> Kefa
>
>
> Bill Ross wrote:
>
> > It would help to send your mdin.
> >
> > Bill
> >
> > _______________________________________________
> > 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


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Aug 19 2009 - 23:21:33 PDT
Custom Search