[gmx-users] PME scalling

[K.A. Feenstra]

Dmitry Kovalsky wrote:
> 
> Hi there
> 
> I have a system with 25000 atoms and want to run a PME simulation on parallel
> I have 4 dual athlons connected with fast ethernet
> and when i setup 1ns simulation at one node (2cpu) it runs 1.5 days with
> perfect cpu loading 99.8% each
> but when I run at all 8 cpu's it runs 2.5 days! with 20% and 40% cpu usage one
> every node.
> is it a problem of network speed?

25k atoms is still a pretty small system; consider that you will only
have 3000 atoms on each node. My guess is that netswork overhead/latency
are killing you here, specifically on the grid (for PME) communication.
If you have a fairly large number of runs you'd want to run anyway,
don't worry on the scaling, simply run them together. Otherwise, you'll
have to wait for David and Eriks long-planned improved scaling 
implementation... ;-) We've discussed this in length a year back in
Groningen; the idea's are there... Any volunteers?

> should I change   fourierspacing and pme_order as Erik proposed
> 
> or It would be better to use ReactionField?
> 
> I'm using near 40 ions in the system

Seeing your large charge density, I'd bet on fourierspacing etc.
and not on reactionfield (that means cut-offs, right? Bad idea
any day, doubly so with all your ions).

If you could come up with a convincing reason not to use all
your ions (preferably none), you could come away cleanly with
a cut-off/rf setup and get the right scaling... Depends on what
you need.

-- 
Groetjes,

Anton
 ________ ___________________________________________________________
|        | Anton Feenstra                                            |
| .      | Dept. of Pharmacochemistry - Vrije Universiteit Amsterdam |
| |----  | De Boelelaan 1083 - 1081 HV Amsterdam - The Netherlands   |
| |----  | Tel: +31 20 44 47608 - Fax: +31 20 44 47610               |
| ' __   | Feenstra at chem.vu.nl - http://www.chem.vu.nl/afdelingen/FAR|
|  /  \  |-----------------------------------------------------------|
| (    ) | Dept. of Biophysical Chemistry - University of Groningen  |
|  \__/  | Nijenborgh 4 - 9747 AG Groningen - The Netherlands        |
|   __   | Tel +31 50 363 4327 - Fax +31 50 363 4800                 |
|  /  \  | K.A.Feenstra at chem.rug.nl - http://md.chem.rug.nl/~anton   |
| (    ) |-----------------------------------------------------------|
|  \__/  | "If You See Me Getting High, Knock Me Down"               |
|        | (Red Hot Chili Peppers)                                   |
|________|___________________________________________________________|