Hi again so,

On 14/05/2012 22:25, Kenneth Hoste wrote:
> If you have suggestions or remarks, please let us know.
[...]
> and can give priority to stuff that would be useful to others

Surely you ask the right person! I've been spending some good time in the past
to find out popular HPC codes among -European mostly- HPC sites; here we go:

* http://eniac.cyi.ac.cy/display/UserDoc/Packages
  - The first line there are the PRACE benchmark applications; this list has
    been decided over many years of work and should be a very good first pick.
  - The next couple lines are runner-ups and other popular software codes.
* http://eniac.cyi.ac.cy/display/UserDoc/Libraries
  - The most commonly used HPC libraries

An alternative way to organize the work would be per software domain:
* http://eniac.cyi.ac.cy/display/UserDoc/LS2_06-01 # OSS Math Libraries
* http://eniac.cyi.ac.cy/display/UserDoc/LS2_10-02 # OSS Language environments
* http://eniac.cyi.ac.cy/display/UserDoc/LS2_11-95 # Molecular Dynamics et al
* http://eniac.cyi.ac.cy/display/UserDoc/LS2_11-97 # Climate Science et al

My suggestion for you now would be the MD codes, because you will arrive
at the success/fail states much faster (==quick progress) and by the end
of it, that community can be served really well (& invited for testing).
Of course, you'll find your self needing to build its dependencies!

In the meantime, I am looking at Bioinformatics software codes
and will come up with a shortlist for that, too.


Also, I'd like to bring to your attention the following notation as regards
the modules namespace: (Note "-<MPILibrary>-<Compiler>-<Precision>" suffix)
http://apps.fz-juelich.de/unite/files/unite-installguide.pdf

Here are some good practices for module names (copied as-is from elsewhere):
* Conforming sites MUST put the Version string after the "/" separator
* Conforming sites SHOULD prefer Version strings with dots notation
* Conforming sites SHOULD define a default Version when multiple ones exist
* Conforming sites MAY further expand the Version string as per local needs
* Conforming sites MAY create subcategories of modules as per local needs
* Conforming sites MUST document their module configuration & namespace policy

There are pros & cons with the different choices on the above,
but I skip it for now since this would make a separate thread ;-)

ps/fyi.
My experience is mostly with RHEL* derivatives but, it happens so, that I now
manage Debian machines; I'm interested to see one solution working all around,
bringing more homogeneity across HPC sites and less clutter.

At your disposal for any further comments/work.

best,
Fotis

-- 
echo "sysadmin know better bash than english" | sed s/min/mins/ \
        | sed 's/better bash/bash better/' # Yelling in a CERN forum

Reply via email to