[aspect-devel] Raw data output

[Walter Landry]

Timo Heister <heister at math.tamu.edu> wrote:
>> What does deal.II use for checkpointing?
> 
> a combination of a custom/efficient binary format to store the
> solution vectors and the distributed mesh and boost:serialize for all
> the data structures in aspect itself (parameters, etc.).
> 
>>  That might be a good choice
>> for the "canonical" output format.
> 
> No, I don't think we can put enough information into a general output
> format so that we can a) snapshot/resume very efficiently and b) have
> this readable for anyone else (for example hdf5+xdmf).

Then it may be good to write a converter to transform the snapshot
info into something Visit can process.  Otherwise, you will be writing
the same data twice during a run, once for visualization and once for
checkpointing.  Gale used to do something like what deal.II does now,
with separate formats for visualization and checkpointing.  It
complicated data management and made data output slower, especially
when trying to figure out why an input file is causing problems.

Cheers,
Walter Landry