[Mitgcm-support] parallelization problem

[mitgcm-support at dev.mitgcm.org]

Dear support team,

I have a curious problem concerning parallelization. I made some
benchmarks with the DOME configuration (450x210x30 grid points, version 
r1b1) on the NEC SX-6:

- using 1, 2 (2x1 or 1x2) and 6 (3x3 or 2x3) processors the model is
running fine (got 2.8 MFLOPS on 1 processor).
- using 4 (2x2) processors the model hangs and is killed by CPU limit. A
short test of only 6 timesteps went through and shows that on 2
processors routine "eedie" consumpts 95% of the time instead "cg2d" as
usual. The same happens when using 2 processors but with 2x2 tiles in total.

Do you have any idea why such a problem would happen at a specific 
configuration and how one would debug that?

-Arne

-- 

Dr. Arne Biastoch

Institute for Marine Research         phone: ++49 431 600-4013
FB1 Ocean Circulation and Climate     fax  : ++49 431 600-1515
Duesternbrooker Weg 20                email: abiastoch at ifm.uni-kiel.de
24105 Kiel, Germany

http://www.ifm.uni-kiel.de/fb/fb1/tm/data/pers/abiastoch.html