Cluster Poll Results (tangent into OS choices)

Andrew M.A. Cater amacater at
Tue Nov 4 16:36:32 EST 2003

On Tue, Nov 04, 2003 at 07:38:30AM -0500, Joe Landman wrote:
<snip of good stuff> 
> >BUT, if the
> >entire HPC community actually worked together to bring about that change
> >it wouldn't be that hard.  Too idealistic?  
> >
> I believe it might be too idealistic.  This crowd, if you read this 
> forum and some of the others, likes to innovate and create its own value 
> atop some sort of standard offering.  If I am reading you correctly, you 
> are advising focusing on making on particular platform that you 
> personally (to separate you from your employer here) like, as the 
> standard, and stop all the bickering about doing another direction (that 
> you personally do not like).  Is this a  fair read?


There is a problem here: the sky is falling, and no-one is listening to 
us :)

Red Hat Linux as we have known it for the past few years has changed
focus.  Most of the Red Hat Linux boxes out there will be unsupported
after December this year - the remainder will be unsupported after April 
2004 (unless they happen to convert to RH Enterprise Linux in one or 
other variant)>

Lots of smaller Red Hat based specialist distributions will also 
potentially feel the knock on effect as software is not updated or you
can't get the base OS on which to build any more.  That's OK - it just
doesn't feel comfortable at this point.  The replacement is still a beta 
- although the beta test period is over, Fedora Core 1.0 isn't out yet.
The new model of Fedora Core / Extra and so on is going to be hard to 
get used to - as is the accelerated speed of change and potential lack
of bug fixes. "Don't fix, upgrade" may be the new model.

There are calls to rip out the proprietary bits of RHEL and build a 
Libre version: that would possibly be unwise - you are still tying your
efforts to someone else's code: this is also likely to be code where
fixes are made relatively slowly on a long timescale and where the 
vendor may have other peoples values in mind: the typical EL customer is
not necessarily the typical cluster owner.

Forking small special purpose distributions is potentially a bad idea.
Rocks/Warewulf/Scyld/Caos(if I'm spelling correctly) are all RH based
- on older code - and all in the same marketplace.


There may be an alternative which will guarantee you code freedom,
won't charge for licenses in any event, won't bow to "commercial" 
pressure and won't restrict your code use/re-use/modification/distribution.
If you want an ultra stable platform to which you can freely contribute 
code and which you can use for any purpose - try Debian "stable".  

The release cycle is long, it's updated relatively infrequently but security
patches and major bugs are fixed.  It won't vary wildly from quarter to 
quarter.  It takes two years between major releases - but the code will 
have been tested for longer and, potentially on more platforms. [It's 
been said that Debian is the test bed of choice for the X Window 
System, for example, because it is made to work on many architectures 
and tends to find obscure bugs]

Debian "testing" is in a state of slow flux.  The name is a misnomer in 
that the code is not necessarily beta quality: it should always be the
"testing" for the next build of the full release such that it's 
perfectly usable from day to day / week to week / month to month and 
should be releasable at relatively short notice.  Occasionally, major 
changes may break stuff for a few days: there is a transition at the 
moment, for example, from KDE 2.x (which was released in stable over a 
year ago) to KDE 3.x (which has been working in "unstable" for some 
months).  Because of incompatibilities, some KDE components may not work 
for a couple of days until it settles down.  For 90% of folk using a cluster, 
that sort of thing may be irrelevant.  Testing is asymptotically 
approaching the next stable release - but won't be released as 
stable until its ready :)

Debian "unstable" - may change fairly dynamically: may break but is 
quickly fixed. Latest bleeding edge software trickles down from here
through a defined process until it reaches testing and then, ultimately,
becomes part of the next stable release. Probably not for clusters -
though I have a toy/evaluation cluster running at work on unstable 
purely for the very latest GCC, for example.

Debian encourages everyone to build specialist distributions based on
Debian.  If the hassle's too much, feed in your cluster packages to 
become part of the main distribution.  There is at least one Debian 
newsgroup (debian beowulf) where clusters are of high interest.
Check out what's already within Debian.  As noted in a previous post,
you may find what you want has already been put in place in the 8000+

This is a purely personal post.  It does _not_ represent my employers
or any other person.

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

More information about the Beowulf mailing list