Re: # of processors

From: Gengbin Zheng (gzheng_at_ks.uiuc.edu)
Date: Mon Apr 05 2004 - 00:41:44 CDT

Hi Neema,

This normally should not happen. It may be interesting to see the timing
on 10 and 14 processors, or even odd number of processors to get a
complete curve.
btw, was there any other job running on the cluster at same time? or check
if there are zombie processes on some nodes that consuming cpu cycles.

Gengbin

On Mon, 29 Mar 2004, Neema Salimi wrote:

> Is there any reason that NAMD would perform poorly with a non power-of-2 # of
> processors? My tests on my 35K atom system on our Linux cluster show
> these timings
> (average of 3 simulations, using the final benchmark timings from each):
>
> 1: 2.16 sec/step
> 2: 1.13 sec/step
> 4: 0.581 sec/step
> 8: 0.301 sec/step
> 12: 0.373 sec/step
> 16: 0.183 sec/step
>
> Thanks.
>
> Neema Salimi
> nsalimi9_at_itsa.ucsf.edu
> Graduate Student - Agard Laboratory
> Program in Biophysics
> University of California, San Francisco
>
>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:18:09 CST