[MITgcm-support] Adjoint model runnning?

Chris Hill cnh at mit.edu
Tue Oct 7 10:26:15 EDT 2003


Hong, 

 Here is what Patrick thinks must be going on.

>The experiement has some extensive recomputations going on,
>which explains the factor 100
>(it has nothing to do with the number of controls).

>The dependency of the code size on nchklev_1
>is perfectly OK (innermost checkpointing loop
>held in memory).

Chris

> -----Original Message-----
> From: mitgcm-support-bounces at mitgcm.org 
> [mailto:mitgcm-support-bounces at mitgcm.org] On Behalf Of Hong Zhang
> Sent: Thursday, September 25, 2003 9:10 AM
> To: mitgcm-support at mitgcm.org
> Subject: [MITgcm-support] Adjoint model runnning?
> 
> 
> The document  says generally adjoint running is about
> 5 times slower that forward model.
> But in my case ( exp0, and only one indenpdent variable),
> it seems 100 times slower.
> The total size of the program (shown by 'top' ) seems 
> dependent on nchklev_1. Here are several cases: nchklev_1=36, 
>  SIZE=31M
>          144,       49M
>          360,	    85M
>         3600,      623M
> but for each case, the Resident Set Size (RSS or RES shown by 
> 'top') is only 9M. Can I load more part of the program into 
> Resident memory so it run faster? Have you ever met such 
> problems? any suggestion?
> 
> thank you,
> hong
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org 
> http://dev.mitgcm.org/mailman/listinfo/mitgcm-> support
> 




More information about the MITgcm-support mailing list