AMBER Archive (2005)

Subject: RE: AMBER: parallel AMBER on cluster

From: Ross Walker (ross_at_rosswalker.co.uk)
Date: Thu Dec 15 2005 - 12:38:57 CST


Dear Katerina

> We recently obtained an access to cluster and want to install our
> AMBER8 there. But system administator doesn't want to install
> AMBER on parallel queue. He told me that the improvement in the
> speed of calculations will be only about 5-10%, because the nodes
> in this cluster are connected by Gigabit network and this limits
> the performance of AMBER. I would like to khow whether
> this is true.
> Any opinions on this subject would be helpfull.

While it is true that gigabit is not a very good interconnect for clusters
that are going to run fine grain parallel jobs the 5-10% quote is way off
the mark. Both sander and PMEMD will work fine on 4 cpus and in most cases
pretty good on 8 cpus.

With the JAC benchmark and gigabit I get around 3.3x speedup for sander.
PMEMD will scale slightly better. On 8 cpus I get around 5.6x speedup.

Your milleage may vary depending on the quality of the network cards and
switches in the machines and whether they are dual cpu nodes or not. If they
are dual cpu nodes then on 2 cpus you can pretty much get 2.0x speedup. GB
runs will also scale pretty well over gigabit compared to the PME runs. Also
the larger your system the better scaling you are likely to get. My advice -
for PME runs use pmemd and you should be able to get reasonable scaling out
to 8 cpus. More than that you will have to benchmark and find where the
sweet spot is. 16 may be faster but in some cases may be slower... For GB
simulations use sander and you should easily be able to go to 8 cpus if your
system is big enough...

Some example timings on a gigabit cluster (I forget the cpu specifics) for
LADH (Using sander):

1 733.10s [1.0x]
2 397.51s [1.8x]
4 207.75s [3.5x]
8 120.41s [6.1x]

I hope this helps.

Ps. When I get a chance I'll try and add some benchmarking numbers for our
3.4GHz dual P4 gigibit cluster to the benchmarks page of the amber website..

All the best
Ross

/\
\/
|\oss Walker

| Department of Molecular Biology TPC15 |
| The Scripps Research Institute |
| Tel: +1 858 784 8889 | EMail:- ross_at_rosswalker.co.uk |
| http://www.rosswalker.co.uk | PGP Key available on request |

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.

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