[gpaw-users] GPAW sprint

[Jens Jørgen Mortensen]

To those of you that are in Lyngby on Monday April 12:

At this weeks GPAW meeting, we decided to have the next GPAW-sprint on
Monday April 12 at DTU in Lyngby starting at 10:30.

The topic of the sprint will be:

  "Testing of GPAW with real life examples"

Yes, we do have a test-suite of many tests, but those are small, trimmed
down and artificial tests.  We need to do more realistic test jobs for
bigger systems that calculate something of value to us.

These jobs will require more CPU's and take longer time to run, so we
will submit the jobs to Niflheim every weekend.

Everybody is invited to participate, and there will be enough tasks for
everyone:

* write job scripts
* develop machinery to submit the jobs to Niflheim
* document how to write tests
* analyze results


Job scripts
-----------

There are already some jobs here:

  https://trac.fysik.dtu.dk/projects/gpaw/browser/trunk/gpaw/sunday

but I would also like most scripts that were used to generate the tables
and figures of the big GPAW paper
(http://www.psi-k.org/newsletters/News_98/Highlight_98.pdf) to be added
as tests.  I have these scripts somewhere, but they need to be polished
and added to SVN.


Submitting the jobs to Niflheim
-------------------------------

I have some code that does that, but it is only a prototype:

  https://trac.fysik.dtu.dk/projects/gpaw/browser/trunk/tools/sunday.py


Documentation
-------------

Someone should describe how one can add new jobs.

I would also like some of the tutorials on the GPAW web-page to be
coupled to jobs that run every weekend.  An example:


https://wiki.fysik.dtu.dk/gpaw/tutorials/lattice_constants/lattice_constants.html#careful-test-for-convergence

The figures in this tutorial take some time to generate, and it would be
nice if this was done automatically every weekend.  This sort of thing
should be made easy and documented somewhere.


Analysis
--------

Once all the jobs have been submitted to Niflheim and have finished we
will need to analyze the results and maybe send an email to the gpaw-dev
ML if something strange happened like a crashed job or if the wall clock
time increased compared to previous weekends (or decreased).


Tasks
-----

* submit machinery (1 person)
* coupling between web-page generation and weekend tests (1-2 persons)
* analysis (1 person)
* writing job scripts (as many as possible)

Please bring your own laptop (or borrow one).  Make sure it has the
right tools installed and test the wireless connection before we start.

If you are in doubt whether or not you should participate - come and
talk to me and I will convince you that you should :-)

See you Monday,
Jens Jørgen