[MITgcm-support] IO on NEC SX machines

Jens-Olaf Beismann jbeismann at hpce.nec.com
Fri Nov 16 07:01:46 EST 2007


Hi,

> The obvious solution is to read forcing in larger chunks (e.g. per year) 
> and keep it in memory, but that would require considerable recoding.

another solution is to keep the forcing data on a disc which is attached 
locally to the SX node - in that case, the IO overhead also is 
negligible, but this is less comfortable than using the GFS.

If this can't be handled by any existing switches, I'd recommend writing 
a new routine to make use of buffered IO on a GFS.

Cheers,

Jens-Olaf




More information about the MITgcm-support mailing list