Hi Carlos,
> we should keep this discussion going- I think you missed out on the
> HREMD brainstorming session at the bar during the amber meeting.
> exchanging coordinates will likely be quite inefficient for many
> problems, but needed for others. we talked a lot about how we need to
> generalize the exchange, and really track down impact on the code (for
> example, needing to redo load balancing and spatial decomposition
> work, etc).
Indeed exchanging coordinates is VERY general but inefficient which is why
we implemented the Temperature REMD first to get the framework in place but
avoid having coordinate exchanges. Obviously we need to do this slowly to be
careful of breaking existing performance in PMEMD and other things as well
such as the REMD changes that went in breaking the GPU code etc. But things
are looking good and Jason's numbers so far for just Temperature REMD look
awesome from a performance perspective. The next step is to implement a
general REMD method but I think we (you, me, Jason and Adrian) should talk
about this offline to determine the optimum way of doing this. Exchanging
coordinates is general but is not going to cut it at all from a performance
perspective, unless you limit exchanges to every 1000 steps or so.
I think one thing we need is an exact definition of what information is
needed for which type of exchange. This needs to be detailed and represent
the minimum amount of information that would need to be exchanged. For
example just saying exchange coordinates is easy but really doesn't help
solve the problem. So if someone wants to volunteer to document specifically
which type of exchanges should be supported and what the specifications for
each type of exchange are, along with working out the minimum information we
can get away with exchanging that would be very helpful.
Probably best we discuss this between a few of us over skype to avoid lots
of confusion, although having a side discussion on here is probably useful
as well.
All the best
Ross
/\
\/
|\oss Walker
---------------------------------------------------------
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Adjunct Assistant Professor |
| Dept. of Chemistry and Biochemistry |
| University of California San Diego |
| NVIDIA Fellow |
|
http://www.rosswalker.co.uk |
http://www.wmd-lab.org/ |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
---------------------------------------------------------
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Fri Apr 08 2011 - 09:30:05 PDT
