Firefly and PC GAMESS-related discussion club

 
Learn how to ask questions correctly  
 
 

 

Alex Granovsky
gran@classic.chem.msu.su

what is the active space used in your MCSCF calculations?
Provided it is not very large, the most time-consuming part
of calculations is the CASSCF integral transformation.
With Firefly, it is fast and scales reasonably well.
We are routinely using Firefly for similar calculations using
ca. 96 to 256 cores. We typically use mvapich version of Firefly.

The hangs you described looks like MPI-related problems.
What is the MPI implementation you are using? Is it OpenMPI?
something else? Does it really run over Infiniband?
If it does not, this could explain a limited scalability.  
Where the Firefly's temporary files are located?
What is the version and build number of Firefly you are using?

Kind regards,
Alex Granovsky


On Wed Mar 6 '13 1:00am, Ilia Solov'yov wrote
---------------------------------------------
>I have a question about Firefly parallelization. I am running
>MCSCF calculations for a system of about 150 atoms.

>I was wondering what is the reasonable number of computer cores I could
>use to get the best performance.

>I have access to large supercomputer cluster where I can use up to
>1000nds of cores for the computations, but as far as I can judge if I
>use more than 30 cores the program becomes very unstable (hangs
>periodically without producing any output), and I do not get any speedup anyway.

>Could you please advise how I can get the maximum performance through
>parallelization.

>Many thanks in advance.