Re: [LUG] Linux Clusters!!!!

[Simon Waters <Simon at wretched.demon.co.uk>]

Jon Still wrote:
> 
> Mmmm....cluster...
> 
> There are plenty of solutions available which quite often don't need existing
> software to be adapated to take advantage of the cluster - unfortunately it
> seems that they are generally quite intensive on network usage, to the point
> where 10Mbps isn't sufficient.

What are you going to do with the cluster, as different
people mean different things by clustering...

If you mean build a compute cluster - which most people seem
to take, then Jon is right, you can do this at a more
abstract level than the OS - witness Seti at Home - and for
most interesting problems the interconnect rapdily becomes a
bottleneck - look at what Cray wire T3E nodes together with.
There are PVM tools for most (all?) Unix systems to solve
maths problems.

You'll have to cluster a lot of 486's to out compute my
Pentium laptop (let alone a G4 Qube) *8-)

More interesting I think is clustering for availability,
figuring out how to make an application that is not
trivially read only (or read mostly) available when one
piece of hardware breaks...

Also how does each part of the cluster know if the other is
dead, and how do they resynchronise after a breakup...
Ideally such clusters should run faster when everyone is
working, and degrade gracefully....

For this second two or three 486 PC's are perfectly adequate
to demonstrate ideas, prove concepts.....

Whilst clustering for a relational database sounds fun, how
about starting with something simpler, like a mail server or
a fileserver. SMTP mail servers are useful, they bottleneck
quickly (many PC based ones can only handle 20 msgs a
second), and reliability is critical.

The first issue I guess is how to make the storage suitably
redundant.
--
lug-list - The Mailing List for the Devon & Cornwall LUG
Mail majordomo at lists.termisoc.org with "unsubscribe lug-list" in the
message body to unsubscribe.