Re: NAMD distribute

From: Shantenu Jha (s.jha_at_ucl.ac.uk)
Date: Thu Oct 13 2005 - 18:08:56 CDT

Dan,

I gather that your motivation is to use "free CPU cycles" for simulations
using NAMD.

If so, probably the easiest (optimal?) approach is to use NAMD in a Condor
environment. That is probably the quickest solution for "cycle stealing".
Just set-up a "Condor Grid" -- if your university doesn't already have one
-- and you're almost ready to roll. Condor takes care of resource
allocation, dynamic management and (dare say) even security.

> > > I have a bit of an odd idea, but it struck me as I walked past a bank of
> > > computers all running Folding_at_Home. Has anyone done anything towards
> > > integrating NAMD with distributed grid computing methods? While this is
> > > clearly feasible, is it useful?

If you want to scale beyond Condor grids the issues that need to be
addressed change. Quickly and drastically.

Our group has been using NAMD on grids for at least two years now. The
grids in question though are not "aggregation of single distributed
processors" but "federation of supercomputers".

There are reasons why we haven't invested time in integrating NAMD in a
SETI_at_home style environment.

 - Why use a heavyweight MD engine like NAMD when you're going to be
   essentially cycle stealing at the level of 1 processor?

  [If I were to ever chose to implement an MD engine at the core of
  SETI_at_HOME or proteinfolding_at_HOME style project, it wouldn't be NAMD. In principle,
  it is reasonably easy to virtualize any code for such a grid
  environment; e.g. United Devices will do it. For a fee, though with
  sufficient charm, and academic clout possibly free...]

 - But more fundamentally, I don't think the approach of distributing
computation at the level of many localized single processors will work
well for *large-scale* simulations. At least not with the current
"legacy" programs, and/or without radically different programming
models/approaches. Or possibly great progress in algorithms...

> So what I gather is that this would be doable if;
> 1) the communication between the grid and server is not significant
> and 2) if the calculations could be secured.

I think the answer lies in the kind of grid you're referring to. There
are grids in existence today that provide you both -- high quality
communication and security.

My $0.02.

Shantenu

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:40:02 CST