Run the model in parallel

[luyssaert]

Try to run the model in parallel for (1) 2, 4, and 9 pixels with 2, 2 and 3 procs. (2) run in parallel for a latitudinal band and (3) for the global domain (2 x 2 degrees).

[aresovsky]

Model runs for a 30-pixel area encompassing most of France. Will try to run for a latitudinal band next.

[aresovsky]

Line 2263 of stomate.f90 is causing the model to hang when run in parallel:

CALL stomate_accu (kjpindex, 1, do_slow, t2m, t2m_daily)

The cause is unclear. This line has been commented out for now (as of rev. 4204).

[luyssaert]

r4204 was tested on Obelix for a 2x2 pixel domain (with a slightly different run.def than in previous tests). The test revealed at least 3 problems: (1) *_stomate_history.nc files were not written, (2) the *_stomate_ipcc_history.nc files contain identical values for lai for the different pixels and no values for gpp, and (3) the model crashed in year 2. the four pixels, however, could be run as 4 serial jobs on 4 different processors (these simulations looked OK) as well as a single multi-pixel job on a single processor (this set-up resulted in equal lai and no gpp for the individual pixels).

[luyssaert]

The problem was caused by an inconsistent use of get_printlev. all get_printlev's were commented out which solved the problem (r5031). Haven't figured it out yet how to correctly use different printlevs in different subroutines of a single module.