Running in parallel mysteries
I want to run a fine mesh (1200*400) on OF and for that I need to run in parallel. I run a first time with some 18cpu and the simulation went to 0.1sec after 3 days. I took the same folder and ran it from the 0.1sec with 36cpu since 4 days, it barely reached the 1.1sec…
I then ran some test cases with 1, 2, 4, 8cpu, during 5minutes for a 150*50 mesh. Here are the results:
1cpu: Time = 0.0569324
2cpu: Time = 0.0680509
4cpu: Time = 0.0638118
8cpu: Time = 0.00469057
16cpu: Time = 0.00384501
Quite strange isn't?
To run in parallel, I simply change the "numberOfSubdomains" in the decomposeParDict file and change the n(x y z) in the simpleCoeffs knowing that x*y*z must be equal to the numberOfSubdomains.
That's it, do I have to change any other parameter?
Thank you for your help, I'm quite new in cfd…
A common mistake is to forget the "-parallel" option with the solver, which can explain why it's slower with more core ....
|All times are GMT -4. The time now is 15:41.|