parallel scaling with sliding interfaces
I have a complex problem with multiple sliding interfaces that needs to be run in parallel.
However, the run time with 2 processors is longer than with one. I have tried different partition methods, solution methods etc. with little change. Does parallel work better on a particular operating system?
Is there any known workaround for this? I have contacted FLUENT, but have not received any useful information to help with this problem.
Re: parallel scaling with sliding interfaces
you will need to give some more information for someone to diagnose this problem.
Important issues are: problem size, physics, models partition methods tried and statistics from partitioning any "special" solution procedures, output monitors, udf's
Think about things that have to be "gathered up" over various partitions at intervals...for example a volume monitor over the whole domain can seriously affect speed in parallel. Do you need it calculated ever iteration? Same with udf's that operate across partitions.
My advice would be to trim the problem back to basics and see how the parallel performance is and then add any of the fancy stuff one by one to see where it deteriorates...then report this back to us/fluent.
|All times are GMT -4. The time now is 02:46.|