[MITgcm-support] speed up!

[Christopher L. Wolfe]

Van Thinh,

I think your code is spending most of its time writing checkpoint 
files. I usually set my chkptFreq to be about a tenth my pChkptFreq. 
That way you don't loose much if the model crashes, but you're not 
spending all your time waiting for the checkpoint files to write.

Christopher

-----------------------------------------------------------
Dr. Christopher L. Wolfe             		 858-534-4560
Physical Oceanography Research Division    OAR 357
Scripps Institution of Oceanography, UCSD  clwolfe at ucsd.edu
-----------------------------------------------------------

On Mar 21, 2007, at 1:17 PM, Van Thinh Nguyen wrote:

> Hi,
>
> I am using mitgcm to simulate internal solitary waves generated by 
> tidal forcing over a topography. So I need to run the model with 
> options: non-hydrostatic & free surface, and OBCs package.
> Everything is doing well with lower resolution, however the 
> computational time is a big concern when I would like to increase the 
> resolution to see the solitary waves.
> I currently used a resolution of 4000*500*60, running on Altix 
> Itanium2 (with 128 cpus x 1.60 GHz), using 32 cpus. To run for 0.1 
> tidal period T
> (4464 s), I need 5 days (time step dt=2s).
>
> I did check the memory when the job runs is:
>
> VIRT = 161 GB (swap + res)
> RES  = 2.3 GB (code + data)
>
> I also saw the job state is running about 60 s with 100% CPU then 
> start to delay about 40s (maybe due to communication between threads, 
> read/write).
>
> Maybe my chkptFreq (=1.) is set to low?
>
> -----
>  pChkptFreq=4464.,
>  chkptFreq=1.,
>  dumpFreq=2232.0,
>  monitorFreq=558,
> -----
>
> Any idea to help me to speedup my run? (switch off some parameters, 
> which is not neccessary in my case? shall i increase chkptFreq?, ....)
>
>
> Thanks so much.
>
> Van Thinh
>
>
> -----------------------------------------------
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
>