Cluster Poll Results (tangent into OS choices)

Alan Scheinine scheinin at
Wed Nov 5 05:22:53 EST 2003

Joe Landman wrote:
 > There are interesting bits in debian.  I am not sure it is necessarily
 > the right choice for clusters due to the specific lack of commercial
 > support for cluster specific items such as Myrinet, and the other high
 > speed interconnects.

The above comment is just one of many that seem to me to describe the
situation as being dependent on a commercial distribution whereas,
in my experience, Red Hat was not sufficient.  The most stable Red Hat
for a long time was 7.3 but to solve a problem with the ext3 file system,
I installed the most recent kernel from  Moreover, from
time-to-time I update gcc/g77/g++ from Gnu.  If I waited for Red Hat,
I would be out-of-date software for the compiler and the kernel, for
a non-trivial period of time.  Perhaps as a slogan I should write: its
not that we need a distribution as good as Red Hat, we need something
even better.  I was motivated to write this when I read the reference
to Myrinet.  The gm driver needs to be compiled with a specific kernel.
Using the most recent kernel (at the time I did the work around the
beginning of 2003) from and using the most recent gcc from Gnu,
I built gm and MPICH-gm and it works fine.
   With regard to the comment by Joe Landman, I assume he is referring
to a cluster-specific distribution such as ROCKS, whereas my comments
make reference to Red Hat.  My intention is to raise a general question,
wouldn't any RedHat-like distribution be sufficient as the base such
that one person could do the rest of the work needed to build and
maintain a cluster?  On our cluster we have MPICH_pgi, MPICH_intel,
MPICH_gcc, MPICH_pgi_myrinet, MPICH_intel_myrinet and most of these
also compiled for debugging.  It not fun to build but I want to give
users a choice.                 ^was
Would I have the same flexibility automatically with a distribution
oriented towards clusters?

Alan Scheinine
Beowulf mailing list, Beowulf at
To change your subscription (digest mode or unsubscribe) visit

More information about the Beowulf mailing list