The OPAL Discussion Forum

[Nicole Neveu <nneveu AT hawk.iit.edu>]

Hi All, 

I have a question about scaling simulations up. What can I do to ensure a run on more cores is faster and efficient? Are there some general changes to the OPAL input file that help with the transition? i.e. SC grid, parallelization, etc.

When trying this on a small test case (16 to 32 cores), so far I only changed my parallelization (based on an old email from Andreas).

16 core run: 

FS_SC: Fieldsolver, FSTYPE = FFT, 

MX = 32, MY = 32, MT = 32,

PARFFTX = false, 

PARFFTY = false, 

PARFFTT = true,

32 core run: 

FS_SC: Fieldsolver, FSTYPE = FFT, 

MX = 32, MY = 32, MT = 32, 

PARFFTX = true, 

PARFFTY = true, 

PARFFTT = false,

I see a 1 minute difference in this case (6 min to 5 min).

What else could I do to help with the walltime?  

Or is this near the limit because the test problem is so small?

Thanks! 

Nicole