[petsc-dev] Monitor changes

[Barry Smith]

On Aug 13, 2011, at 10:31 PM, Jed Brown wrote:

> On Sat, Aug 13, 2011 at 17:47, Barry Smith <bsmith at mcs.anl.gov> wrote:
>   0 SNES Function .....
>       0 KSP ....
>       1 KSP ...
>   1 SNES Function ...
> 
> 
>   In JASON should be
> 
>     KSP   0 stuff 1 stuff ....    (acceptable diff information)
>     SNES 0 stuff 1 stuff ....   (acceptable diff information)
> 
> It may be slightly more complicated with nested solvers and since the number of iterations can be different without necessarily indicating a problem with convergence. In any case, if we can create a normalized "detail" form, then it would be much easier to do monitoring. 
> 
> With a configuration GUI, I would like that context to remain live and have various monitors that you could look at to see how the solve is proceeding at each level. Since this is read-only and (currently) less structured than configuration options, it's not clear to me whether AMS is the best way to get this information out.

   AMS could be ok. It doesn't care about structure. 

> 
> Here are some possible options, not necessarily mutually exclusive:
> 
> 1. Convert monitor/view output to produce some normalized structured representation in C data structures.
> 
> 2. Log everything to a sqlite database (because sqlite is small and ubiquitous and directly supports common queries that people would like to do).
> 
> 3. Write everything to normalized JSON which can "easily" be parsed to recover a queryable form.
> 
> 4. Write the human-readable form as a filter on 1, 2, or 3, or write it separately.