Posted by Evgeniy on November 27, 2009 at 13:43:01:
Dear Developers and Users of Molcas,
I have a question concerning I/O in Molcas, in particular in the seward module. I recently encountered incomprehensible slowing down of seward when running molcas in parallel. In my calculation seward (with CD option) is followed by scf, rasscf and caspt2 calculations. As far as I know all these are parallelized except for the caspt2 module. However when I run the job all modules work pretty fast except for seward which normally should be the fastest one. It takes it even longer to finish than caspt2 does. I observe this on a big cluster which has a global scratch disk. So, the I/O to the scratch disk might be the main bottleneck. But then it should be that for all modules and especially for caspt2 that writes much more on the scratch disk than seward. Nevertheless the former runs faster than the latter. What can be the reason of slowing down for seward? Many thanks!
Best regards,
Evgeniy