Re: Parallelization of a code from TCB Summer School 2003

From: Alex V. (vav_at_udman.ru)
Date: Mon Apr 19 2004 - 02:49:05 CDT

Hi Yang!
Yes, small number of atoms can be the case of slow parallel
simulation. Send/recieve operations consume more time than calculations.
But sometimes it is because of your cluster properties.
Are all of your CPU's of the same speed?

Alex

Yang Yang wrote:

>Hello everybody,
>
>I am learning to use NAMD by starting from one of the TCB summer
>school tutorials, i.e. "Simulating Membrane Channels-Water channel
>tutorial".
>
>After successfully running it on one processor, I am trying to run
>it on 3 and 8 processors. I did not alter any part of the configure
>file, everything remains the same except for the command line.
>
>I used the following to invoke NAMD
>
>"mpirun -machinefile list -np 3 namd2 sim_short.conf", where the
>list is the file containing the processor names on which the namd
>will run.
>
>It did run properly but very slowly. The CPU time is 11404.68 and
>the wall clock is also 11404.68 which is much longer than the time
>consumed by runnning on one processor which used CPU time/Wall clock
>at 3384.39.
>
>Questions, can I simply run it in this way for parallel computing?
>Or is it because the atom number is so small, only 1320, that NAMD
>can not work efficiently?
>
> I will appreciate any advice.
>
>Yang
>
>
>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:18:11 CST