[MITgcm-support] file size issue with mpi

[Dimitris Menemenlis]

On Tuesday 31 August 2004 09:59 am, Uli Riemenschneider wrote:
> Will "useSingleCpuIO" solve the problem, or should I be writing out 
> tiled files?

Uli, on the SGI Origin "useSingleCpuIO" is definitely the way to go.
You'll find that it is way faster and more robust than the "globalfile"
option, and depending on configuration (if you are writing to a single
disk) it can be faster than the tiled option as well.  And you don't
end up with a gzillion files that you have to join a posteriori.

>> Also, my run produced no STDOUT.*.* files, but instead printed the 
>> monitored fields for all processors to the screen in one big muddled 

Not sure why that is.  I have used SGI Origins at both NASA AMES 
(MITgcm/tools/build_options/ames_opt+mpi) and at JPL 
(MITgcm/tools/build_options/o2k+mpi) and have not encountered
this problem.  Maybe something to do with MPI configuration?

D.

-- 
Dimitris Menemenlis <menemenlis at jpl.nasa.gov>
Jet Propulsion Lab, California Institute of Technology
MS 300-323, 4800 Oak Grove Dr, Pasadena CA 91109-8099
tel: 818-354-1656;  fax: 818-393-6720