AMBER Archive (2003)

Subject: Re: AMBER: PMEMD Performance on Beowulf systems

From: David E. Konerding (dekonerding_at_lbl.gov)
Date: Mon Dec 22 2003 - 11:04:03 CST

Robert Duke wrote:

>David -
>Thanks much for the specific inputs on good vs. bad cluster hardware. I
>have observed differences between cheap and more expensive hardware, but not
>systematically. In general, I would trust server grade setups, including
>switches, more than the average desktop pc, which can be a piece of junk
>from an i/o perspective. However, specific recommendations are more useful
>than general guidelines. PMEMD is designed with computation and i/o overlap
>in mind; but not all hardware will take good advantage of it.
>Regards - Bob
>
>

Hi Bob,

The one lesson I've learned is that it's a lot harder to evaluate PC
hardware than I ever imagined. I agree
completely about the server boxes tending to have better I/O
construction, but these days, I leave the hardware
evaluation to our local gurus. They tend to notice interesting things
like how some newer machines,
with faster processors, have slower I/O than some older machines with
slower processors, because the ratio of the CPU bus and the PCI bus is
larger- fewer
actual cycles in which the CPU and PCI can trade bytes. Our local guru
also correctly noticed that the newer Intel chips (which really aren't
that expensive, and can be found on the motherboard if you look closely-
make sure they hang the chip directly off the motherboard rather than
placing it "behind" the PCI
bus- this makes some difference.

Sometimes, if you have a local guru who can identify cheaper hardware
that performs well, it's a win to listen to their recommendations.
Clearly, caveat emptor applies in this case.

Dave

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