Some initial discussions of a Fedora HPC SIG

[Shawn Starr]

Hello everyone,

There is talk of creating a HPC SIG that will focus on making Fedora HPC/Grid/Cluster ready. It would be great to find out who in Fedora is interested in beginning some discussions. 

Here's some of my own questions to add to the mix:

- How can we make Anaconda support different interconnects?

- How do we handle HPC/Grid needs which differ from traditional systems? Using puppet, others?
   * Should we provide 'sane' defaults needed within a HPC environment out of the box? 
   * Addon packages to modify configurations of other packages?

- Which job submission/batch schedulers? more choice is better.

- Full OFED integration (Doug Ledford has been working on this)

- Provisioning: Multiple methods for provisioning? 
  * Cobbler, Spacewalk, other open source mechanisms?

- Packaging of HPC applications/libs
  * Open MPI, MPICH1/MPICH2?, MVAPICH1/MVAPICH2?
  * Directory structure - FHS
  * Using environment-modules or alternatives to allow users to switch between different MPIs etc.

- Package Optimization
  * A lot of people tend to want highly optimized packages such as ATLAS and usually will recompile them against their
     new processors. Do we just settle on defaults, can't satisfy all people.

There are many more questions and getting some initial discussions going will be beneficial to Fedora.

--
Shawn Starr
Software Developer, Open Source Grid Development Center (OSGDC)
Platform Computing
3760 14th Avenue
Markham, ON L3R3T7
direct: 905.948.4229
http://www.platform.com