AMBER Archive (2007)

Subject: Re: AMBER: replica exchange

From: Robert Duke (rduke_at_email.unc.edu)
Date: Sun Oct 28 2007 - 10:38:04 CST

Hi Geoff -
PMEMD does not support replica exchange or multisander at present, for several reasons:
- there are additional challenges to scaling multisander and the stuff it supports, that will limit the high end scaling that can be achieved,
- the techniques already have somewhat the characteristics of a collection of small jobs, so sander does sort of okay on them,
- development cost and other higher priorities.
So you'll need to look at these problems in a multisander context for a while...
Best Regards - Bob Duke
  ----- Original Message -----
  From: Geoff Wood
  To: amber_at_scripps.edu
  Sent: Sunday, October 28, 2007 12:29 PM
  Subject: AMBER: replica exchange

  Dear Amber users,

  I am interested in utilizing a Blugene/L machine that I have available to me. I successfully compiled and ran PMEMD calculations on the BG/L using the config files available from the following link http://amber.scripps.edu/config.bgl.html however, I have just realised that remd is not implement in PMEMD. Is there a way of including the routines in multisander in the PMEMD compilation or compiling sander from scratch using similar config files? I don't think sander will have scaling problems on a BG/L when using remd because although you could utilize many cpus it is the number of cpus per replica that counts.

  Kind regards

  Dr Geoffrey Wood
  Ecole Polytechnique Fédérale de Lausanne
  SB - ISIC - LCBC
  BCH 4108
  CH - 1015 Lausanne

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu