Slow down on local parallel
I have a question about parallel performance. Im using a quadcore processor with only 2gb of memmory running on win xp. I have set up a 2D model and the solver slows down the more cores I use when I run in local parallel mode. For me it is a lot faster to run in serial mode. Is this because the bottle neck is the communication between the different partitions or is there another explanation?
Re: Slow down on local parallel
If it is a small grid, the communication overhead may be greater than the parallelization benefit. That said, if you're running with PVM on Windows you may see better performance by using MPICH (or MPICH2 on XP64) instead.
|All times are GMT -4. The time now is 11:40.|