Threading vtk output?

[Konrad]

Dear deal.ii community,

I wrote a shared memory parallel code for a single machine and I am very happy that I can easily compute around 50 million+ DOFs in reasonable time for a stationary problem. However, writing such an amount of data is a bottleneck. I know about distributed memory solutions for writing output efficiently but

is there also a good solution solution for shared memory parallelism? I would like to write vtk-files.

Maybe I missed it while browsing the documentation. Can anyone please point me to the right spot?

Best,
Konrad

[Wolfgang Bangerth]

Konrad,

> I wrote a shared memory parallel code for a single machine and I am very
> happy that I can easily compute around 50 million+ DOFs in reasonable
> time for a stationary problem.

Nice!

> However, writing such an amount of data
> is a bottleneck. I know about distributed memory solutions for writing
> output efficiently but
> is there also a good solution solution for shared memory parallelism? I
> would like to write vtk-files.
>
> Maybe I missed it while browsing the documentation. Can anyone please
> point me to the right spot?

There is nothing there right now, but let's measure first. Have you
timed the different parts of your program? (Like in step-32 or step-40,
for example?) As a percentage of the overall run time of your program,
how long do the operations
data_out.build_patches()
and
date_out.write_vtu (...)
take? These are the two where substantial work happens.

Best
W.


--
------------------------------------------------------------------------
Wolfgang Bangerth email: bang...@colostate.edu
www: http://www.math.colostate.edu/~bangerth/