Parallel efficiency and speedup info
Could you share a little about your parallel efficiency and speedup? I just want to know what's going wrong with my case set-up.
I have read pyFoam, http://openfoamwiki.net/index.php/Be...ks_standard_v1. But I want to know more.
Mine is a flow past a circular cylinder, grid number 860000, mpich1.
And the speedup is far from linear, using 8 processors, the speedup is only about 3.5, and using 16 processors, the speedup is only 9.1, ie. a sublinear speedup of 43.1%. It's very very bad, isn't it?
Any information and comments are welcomed.
I think it may depend on your hardware - I have seen people reporting bad scaling w. ethernet interconnection, but good scaling e.g. using Infiniband.
Here an example:
We also get rather bad scaling currently w. ehternet, and are now installing infiniband.
I cant await the results :)
Ya, I just want to know what's the best results with openfoam...:D
wow, they got superlinear speed up using 64 processors!
|All times are GMT -4. The time now is 14:01.|