[Mitgcm-support] RE: MIT GCM links

[mitgcm-support at dev.mitgcm.org]

Hi Dimitris,

 Thanks. I looked at Gerhard's stuff. I don't think you need to have the
high-frequency textual I/O he noted as one performance bottleneck. Its
probably printing out stuff like solver tolerances and KE every time
step and that could be turned off except for debugging. Also there are
lots of options for binary I/O (as you are all too painfully aware!). If
Gerhard can give us a recipe for what works best on halem I think we can
figure out a way to fix things up quickly.

 Finally the solver scaling will then show up! I believe that switching
the solver communications to shmem would fix this. If you can get
someone to give you time (a couple of weeks) from Gerhard for this that
would be great. I know Gerhard from some other work (ESMF) and he should
be able to do this quite well.

Chris


> -----Original Message-----
> From: Dimitris Menemenlis [mailto:menemenlis at jpl.nasa.gov] 
> Sent: Wednesday, May 28, 2003 4:49 PM
> To: Chris Hill
> Cc: support at mitgcm.org
> Subject: MIT GCM links
> 
> 
> 
> Chris, a few more MIT GCM links.
> 
> Sheila's O3K update:
> 
>     http://people.nas.nasa.gov/~faulkner/NEWMITgcmUV.html
> 
> and Gerhard Theurich test on Compaq cluster:
> 
>