dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

The DOLFIN benchbot is now up and running. Thanks Johannes for a good
effort.

The benchbot runs the set of benchmarks once every night and reports
the results here:

  http://www.fenics.org/bench/

Johannes has installed a range of old DOLFIN versions and backported
the benchmarks to run on those versions.

The benchbot runs all executables named 'bench' found under the bench
directory in DOLFIN. I have added an initial set of benchmarks but
some work remains. What needs to happen now is

1. We finalize on a good set of benchmarks. Some obvious things (like
assembly over different forms) are missing, and the other benchmarks
need to be scaled properly. For example, what is an appropriate
running time to aim for? 1s? 10s? 30s? 1min? The benchmarks that are
there now are very small and run fast (which was useful when we wrote
the scripts).

2. Johannes backports those to earlier version where possible.

Since (2) involves some manual work, the benchmarks need to be
finalized so we don't change them again tomorrow (or next year). We
can always add new (and then the benchbot will start recording new
results as they appear) but changing existing benchmarks will
invalidate all data built up by the benchbot.

Please have a look inside the bench directory and comment on what
needs to be done (or fix it).

Johannes, could you fill us in on exactly what hardware/operating
system the benchbot is running on? It would also be good to add this
information to the web page.

--
Anders

PS: The benchbot page is not themed at all (in addition to the theming
of fenics.org being completely broken at the moment as a result of the
Simula IT department being away on a "seminar"), but it will be nicely
themed in the near future. Harish is working on a new design for the
FEniCS web pages and I can promise it will look good. Very good.

Attachment: signature.asc
Description: Digital signature