TURBOMOLE Users Forum

TURBOMOLE Modules => Aoforce and Numforce => Topic started by: basklau on March 06, 2009, 10:04:41 AM

Title: parallel NumForce no speed up
Post by: basklau on March 06, 2009, 10:04:41 AM

Hi everyone,

with the help of the forum I managed to get the parallel version of NumForce started. But now I realised that there only a relativly small speed up. The problem that occurs is the following:
We have computers with 4 processors. In the hostfile I wrote all 4 processors and it started normally. But after a while all four processes no matter whether dscf or egrad is started have only a load of 0.33  but the overall load is 4 !
I think this is due to file writing. Is there a possibity to reduce filewriting and to use more RAM for example. I cannot find anything in the manual.

Thanks basklau

Title: Re: parallel NumForce no speed up
Post by: christof.haettig on March 27, 2009, 02:46:41 PM

Of cause the I/O load and thus the precentage of time spend in the I/O increases the more processses are started on a single machine. But what you write below sounds for me a bit extrem... How long do the individual points (one dscf or one escf
calculation take?)

Christof