[cam-users] Lack of parallellisation of dynamical package ?

[\yvind Seland]

I am working on aerosol modelling in CAM, previous in CCM3,
and noted that when I used the finite volume dynamical package
in the model does not seem to calculate this part in parallell, neither
in orginal version, nor when including the aerosols.

This does not have much effect with only water vapor as a transported
species, but with 20 species and transport only done in the master
processor, it means a huge increase in total calculation time even
though the increase on the master is not that large.
An example: Basic model 2 days run on a 2x2.5 degree, SGI Origin using 8
processors
Master processor: 486 seconds
Slaves: 190-260 seconds

Total CPU: 3520 seconds

3520/8 = 440 seconds, close to master

Aerosol model: 18 species + some chemistry and deposition which mainly
affects the tphys* routines of the model


Master: 791 seconds
Slaves: 350-390

Total CPU : 5994  (8800 for 16 processors !)
5994/8 = 699, close to master


Is this really the case or have I goofed up the open-mp parallellisation
in some way?



Øyvind Seland          oyvind.seland@geofysikk.uio.no
Post Doc
Department of Geophysics
University of Oslo

Tel: +47 22 85 58 27
Fax: +47 22 85 52 69