AMBER Archive (2008)

Subject: Re: AMBER: Performance issues on Ethernet clusters

From: David A. Case (case_at_scripps.edu)
Date: Thu Apr 17 2008 - 14:02:45 CDT

On Thu, Apr 17, 2008, Sasha Buzko wrote:

> I've just completed setting up pmemd with mpich2 to test on a cluster
> with gigabit Ethernet connections. As a test case, I used an example
> from an Amber tutorial

Why not use one of the standard benchmarks such as jac or factor_ix? They are
there for the purposes of doing exactly what you are talking about, and you
can directly compare your results with those posted on our website for similar
machines.

Once a baseline is established, one can look at other types of jobs.

> Is there anything in the application/cluster configuration or build
> options that can be done (other than look for cash to get Infiniband)? I
> hope so, since it's hard to believe that all the descriptions of
> Ethernet-based clusters (including this one:
> http://amber.scripps.edu/cluster_info/index.html) are meaningless..

The problem has been described many times on this archive, but perhaps the
above page has so much bit-rot it should be taken down. What the above
Stony Brook page describes is a cluster with 800 Mhz Pentium III nodes.(!)
Since that time, computing speed has gone up much faster than interconnect
speeds, so the communication to cpu ratio is much lower now (with gigabit
ethernet) than it was in the old days.

...hope this helps....dac

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