On Wed, Feb 15, 2012 at 10:36 AM, David A Case <case.biomaps.rutgers.edu> wrote:
>
> Aside: is there (or is there going to be) an MPI version of cpptraj?
>
Short answer: No (at least not for this release).
Although there is some MPI-related code in cpptraj, right now it is
unused/disabled. Part of the reason is that parallelization over
trajectory reads proved to be a mixed bag in ptraj; decent speedup for
some things, but sometimes there was no gain to even a slowdown
depending on the cost of the actions being executed, filesystem
performance etc, not to mention it made bookkeeping for things like
data and trajectory writes a pain. So rather than do that I focused on
trying to make some of the more expensive actions (where trajectory
reads tend not to be the bottleneck) faster. It's much easier to do
this with openMP pragmas than with MPI calls (at least in my opinion),
and since many machines come with at least 4-8 cores these days you
can get decent speedup with a shared-mem approach. I don't see there
being any real advantage to using MPI code for analysis code right
now. Maybe for the next AmberTools release (after the Mayan
discontinuity...)
-Dan
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Wed Feb 15 2012 - 10:30:02 PST
