[Beowulf] Clusters and Distro Lifespans
landman at scalableinformatics.com
Wed Jul 19 05:58:24 PDT 2006
Gerald Davies wrote:
> Hi all,
> I'm new to posting, however, i do read the list :)
> John's comment about the short support lifetime of FC raises one of my
> concerns about distros and cluster set-ups in general. In my
> department we have RH/FC based clusters. When purchased they came
> with a pre-installed distro and have pxeboot/images. I then spent
> time tuning them to our needs.
Ok. FC as a distro is supposed to be experimental. Some releases of it
are actually quite good. FC4 has been great on one of boxes (FC5 has
been a disaster on the same box, which has followed a pattern we have
observed of FXe being good, and FCo being garbage, where e is an even
number, o being an odd number).
> My questions relating to this are:
> i) Is the practice of buying clusters with pre-installed distros popular?
The way I look at things, the distro is entirely dictated by the
apps/usage pattern. If you are going to run code which requires
commercial support, with very little exception, the FCx's won't be
supported. With the major distros (RH SuSE) this is not an issue. Of
course the major distros come with some of their own issues; RH using
ancient but supported kernels, and missing lots of important features
and drivers, SuSE using more modern kernels and having lots of drivers,
but having a rather odd way and annoying way of installing some new
drivers. You can use derivative builds of RH, and some of them add to
the base, but most are simply rebuilds w/o copyrighted material to
enable them to be freely used/distributed.
This of course assumes that you will be physically installing the
operating system per compute node. This does not need to be true BTW.
> ii) Would it be better to develop our own installation process for
> clusters so that upgrades, in terms of distros, can be rolled out
> easily? I feel like i'm tied in some way to the supplier of our
> cluster for upgrades.
Hmmm.... you would be re-inventing this wheel, which has been
re-invented many times. Some folks have frameworks set up for installs,
some more tightly integrated into the base OS than others. If you can
live with the RH limitations, Rocks gives you an easy install method.
If you can't live with that, Warewulf gives you a very nice diskless
method to use with a number of distros. Oscar is about, and Bernard Li
keeps wanting us to switch.
We have a framework that works well with SuSE (9.x->10.x), and it should
work well with any other distro, diskfull or diskless, but we have been
working on seeing if we can make Warewulf work with SuSE. Its been hard
as most of the frameworks have lots of built-in dependencies upon their
development distributions, WW depends upon a working yum and a
dependency tree that is very RH centric.
> iii) Do people regularly upgrade their clusters in relation to
> distros? I guess this is like asking how long is a piece of string
> because everyone's needs are different.
Cluster upgrades are rare unless you are missing functionality or
something is broken. That is of course one opinion, some here do
upgrades nightly. From a purely production oriented viewpoint, where
downtime == lost money for our customers, we usually advise against that.
> Apologies if this sounds like a strange first post :)
Its a good post, ask away, lots of folks here would be happy to help.
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax : +1 734 786 8452 or +1 866 888 3112
cell : +1 734 612 4615
More information about the Beowulf