[Cluster-numbers-devs] ClusterNumbers code online!

[Raúl (N. G. <nac...@gm...>]

Hello list!!!

This is the first real post on Dev's list and I'm proud to announce that the
first public release of code is on the GIT repository! You can access it
using your preferred GIT tool using this link (ro):

*git://*
cluster-numbers.git.sourceforge.net/gitroot/cluster-numbers/cluster-numbers

If you want to be an official contributor, please reply to this email
stating so and I'll give you access for committing to the project.

At this point you may be thinking, what is the current state of the tool?

First you need to know a brief history of ClusterNumbers. Back at Q4 2007, I
was finishing it in order to present the tool as my thesis project at my
university, it was pretty stable and functional on x86_64 cluster using
ethernet interconnection and NFS filesystem. All went OK at the presentation
and I got graduated. At this point I had to start working full time, so
ClusterNumbers was set aside for a while.

Now, about 3 years later, I found that PythonCard (toolkit used for the GUI
module) is no longer maintained, a few bugs due to new releases and
deprecations in some Python modules and that Twisted-conch (used for network
programming) is behaving funny (not connecting the GUI with the server
properly). So I propose a few things here.

First of all, I found a possible and simpler replacement for Twisted-conch
called Paramiko <https://github.com/robey/paramiko/>, the main project page
says the last release was done in November 2009, but I've found an Ubuntu
specific page <https://launchpad.net/ubuntu/natty/+source/paramiko> that
seems to be releasing bug-fixes and some improvements for the project. I
already wrote to Robey Pointer, the creator of this tool, asking about the
status of the project. As soon as I get a response from him I'll let you
know.

Second, the GUI is still stable to some extend, I've test it on OpenSUSE
11.2 and 11.3 (both 32 & 64bits), Windows XP and I'm on my way of testing it
on Windows 7 (your help is greatly appreciated on this). I think migrating
from PythonCard to wxPython is the way to go, since PythonCard is based on
wxPython and I haven't seen a reliable fork of PythonCard, but I want to
hear feedback from Devs/Users about this particular point. I think this
could be delayed until having ClusterNumbers up and running again.

Third and last for the moment, the routines that detects the CPU type/model
are in "filesystem/scripts/DetectHW.py:cpu_info()" in the GIT repo, they are
outdated too, but its really simple to add support for new CPUs. It reads
the data from /proc/cpuinfo file and uses RegExp to find a match and set the
correct compiler parameters for the benchmarks. It is based on an idea I
took from the NumPy <http://numpy.scipy.org/> project, specifically from the
file "distutils/cpuinfo.py". Maybe we can reuse some of its code in order to
add support to different Operating Systems, like different Unix flavors and
even Windows HPC.

This is a really short list of improvements needed, there are still some
intra-code comments and messages to translate and of course things that are
actually wrong but you (in this case I) got used to see them and goes
unnoticed.

Thanks everyone for your interest in ClusterNumbers, I hope this tool will
grow with your valuable help.

Best Regards,

*           **
*
*Raul (Nacho) Gomez
Linux Counter #156439*