[gmx-users] scaling

[David van der Spoel]

On Fri, 10 May 2002, Lynne E. Bilston wrote:

>I'm running the dppc benchmark on my new systems, and have found that it 
>takes longer to run the benchmark if I run it over two dual cpu machines 
>than running it on a single dual cpu machine, although the logfiles report 
>76% scaling and 99% scaling respectively. I'm using dual athlons with 100 
>Mb ethernet, and RH7.2.
>
>I assumed this means that the communication between the two machines is the 
>bottleneck. (I'm using an old switch while I wait for the new one)
>
>However, in the logfile there are some comments about load balancing 
>reducing performance. Is this something to look at?
Have you tried using:

grompp -shuffle -sort

furthermore you want to compile LAM with the option
--with-tcp-short=524288

(see http://zorn.bmc.uu.se/~spoel/scaling/#tcp)


Groeten, David.
________________________________________________________________________
Dr. David van der Spoel, 	Biomedical center, Dept. of Biochemistry
Husargatan 3, Box 576,  	75123 Uppsala, Sweden
phone:	46 18 471 4205		fax: 46 18 511 755
spoel at xray.bmc.uu.se	spoel at gromacs.org   http://zorn.bmc.uu.se/~spoel
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++