[MITgcm-support] problems with the size.h file

Wed Nov 6 08:41:32 EST 2013

Hi, please look at figure 5 in http://mitgcm.org/pdfs/96JC02776.pdf
and accompanying discussion.

Then try rereading 2.11.4
http://mitgcm.org/public/r2_manual/latest/online_documents/node47.html

and looking at the SIZE.h example under, e.g., 3.12.3.7
http://mitgcm.org/public/r2_manual/latest/online_documents/node132.html

npx and npy allow you to run the model on a parallel computer using Message Passing Interface (MPI)
specifically you will spawn npx*npy processes, one process per CPU or core

nsx and nsy also decompose the domain in tiles, except that these nsx*nsy run under same process.
there are several reasons for this second decomposition.  One is that it allows you to debug some aspects
of the tiling on a single processor, another has to do with allowing you to run the code in shared memory
configuration, and a third is that sometimes you want small tiles so you can discard the ones that are
100% dry.

for simplicity, you can set npx=npy=nsx=nsy=1 for time being and run small tests in a single-process
configuration. when your problem grows in size, then you will want to explore using MPI.

Dimitris Menemenlis

On Nov 6, 2013, at 4:40 AM, 李志远 wrote:

from :allenson3   <oceanlizy at 163.com<mailto:oceanlizy at 163.com>>
subject :  problems with the "size.h"  file

as can be seen from the manual that , in the size.h file  , two kinds of  parameters as fellow:
 1 ,snx=90,sny=40,olx=3,oly=3 ,nsx=1 ,nsy=1 ,npx=1,npy=1 ;
2 ,snx=45 ,sny=20,olx=3,oly=3,nsx=1,nsy=1,npx=2 ,npy=2 .
I want to know that , the total grid number in X direction  is   snx*nsx*npx  ,in  Y  is sny*nsy*npy   ,  in the above two cases ,the total grid number are the same  , then  what is the difference between  the two cases ?  and , the nsx=1 ,or nsx=2 ,will  cause what different influence ?   as well as "the npx =1 or npx=2 "  ?
someone can tell me what the “tiles” ，“process” exactly mean  ? I have read the manual  many times ,but I 'm still not clear about it .

many thanks !!!

_______________________________________________
MITgcm-support mailing list
MITgcm-support at mitgcm.org<mailto:MITgcm-support at mitgcm.org>
http://mitgcm.org/mailman/listinfo/mitgcm-support

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20131106/220d15ba/attachment.htm>