Hi There,
I’m just curious about set up and efficiency of the parallel version of mmaps. I have some calculations that are really hard to found the best structures, and I want to set up mmaps with the best amount of processors. I had made some tests, but I didn’t find a way to get a good felling of the limits of the parallelization and I don’t want to set up lots of nodes and (maybe) run it in a not good situation.
Anyone have tested this?
Thanks
Gustavo
An alternative is to use the faster enumeration algorithm I’ve recently implemented.
See
https://www.brown.edu/Departments/Engineering/Labs/avdw/forum/viewtopic.php?f=4&t=269&hilit=speed
The parallelization is over different supercells so you won’t get any gains if the number of processor exceeds the number of supercell for the current volume being generated. Now, the number of supercell keeps increasing, so eventually this condition always becomes true, but it may take time.
A good way to try things out in a more controlled setting is with the genstr command, which is also parallelized.