[MITgcm-support] Optimizing the 2D pressure solver

[Van Thinh Nguyen]

Hi all,

Regarding your discussion on scaling the code, I wonder I should provide 
some info what I did, you may interest in this.

I did scale my run with the domain: 4000*500*60 for internal wave 
simulation (free surface, non-hydrostatic) on shared and distributed 
platforms, as follows:

shared : Altix 4700,CPU Itanium2 1.6 GHz, interconnect NUMA

distributed: HP cluster, CPU 2xopteron 2.6 GHz, interconect Quadrics Elan4

Here are some results: compared the runtime in one timestep (=2sec)

    Number of cpus:   64           80          128
    shared           50s/step     30s/step     190s/step
    distributed      40s/step     38s/step      25s/step

I used:

  cg2dMaxIters=1000,
  cg2dTargetResidual=1.E-13,
  cg3dMaxIters=400,
  cg3dTargetResidual=1.E-13

Matt is right, it doesn't make sense if we use too many cpus for a domain 
not large enough, the communication between interior overlap cells will 
make the code runs worse.


Van Thinh



----------------------------------------------------------

On Sat, 22 Sep 2007, Michael Schaferkotter wrote:

> greetings;
>
> i looked over the preceding threads for this topic and found no mention of tile sizes or overlap size.
> aren/t  these specified in code/SIZE.h?
>
> so how is the domain being decomposed?
>
> michael
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
>