[MITgcm-devel] jfnk

[Jean-Michel Campin]

Hi Martin,

I would try just to write this in the STDOUT first, with some prefix
which is easy to grep. This way you can easily get those numbers
and even plot the time series. Later on, can be part of the
seaice monitor (if you prefer this way, or if it turns out to be
more convenient), but would need to be stored until seaice_monitor 
is called and write them.

Cheers,
Jean-Michel

PS: did not check yet the residual calculation, but it's on my list.

On Tue, Nov 06, 2012 at 06:43:05PM +0100, Martin Losch wrote:
> Hi Jean-Michel,
> 
> I would like to compute and a couple of statistics about the JFNK solver, such as
> mean number Newton iterations and Krylov iterations per time steps, number of failures of convergence etc.
> It's very cheap to count these, but what do you suggest as an output scheme? Should they be part of seaice monitor? Or should I just dump these numbers to stdout at, say seance monitor frequency?
> 
> Martin
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel