-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On Friday 10 January 2003 09:56, Richard Jenniss wrote:
> Hello CLUG'ers.
>
> I'm looking for some information on clustering.
> I am not entirely sure where to start.
>
> I assume there's different ways to cluster? or
> Maybe different protocols or methods.
there are two basic types: high performance (HP) and high availability (HA).
HP clusters are all about solving problems that break down nicely into
smaller problems quickly (e.g. fluid dynamics, as seen in the atmosphere or
nuclear explosions).. .they don't do much for NP complete problems, though.
for those you just need Really Big Machines (like the old-style Cray type
super computers). in the linux world, HP is often achieve using Beowulf
clustering and it's MPI framework. the programs need to be written for this
sort of system to see any realy benefits, though. i know sebastian has set up
a beowulf or two in his time =)
HA is for splitting loads between like-configured systems, so that if one
system dies on you it doesn't bring down the entire service and so that you
can (theoretically) scale horizontally (which is far cheaper than vertically,
as you don't have to scrap your current investment). most apps work just fine
in these sorts of environments, assuming whatever data you have can be
replicated across all the machines (or at least accessable). in the linux
world, the two main contenders in this area are Linux Virtual Server (LVS)
and OpenMosix. they tackle the the HA problem in quite different ways and are
therefore useful in different situations.
but it sounds like you are more interested in HP than HA clustering. what are
you interested in doing with it?
> as crazy as this sounds, I'd like to have the cients boot off of a bootable
> Linux CD.
>
> The one host node? I guess, or controller, will have a hard drive, its some
> old equipment left at school. A SPARCserver 1000.
this is actually how most HP clusters work, so it isn't crazy at all. i mean,
do you really want to manage the moving, spinning and therefore often
breaking down hard drives on a 1024 node cluster? ;-)
usually you just want the RAM and CPU resources anyways... the data is often
huge and therefore kept on dedicated disk arrays.
usually the nodes boot over the network, much like diskless clients in LTSP.
- --
Aaron J. Seigo
GPG Fingerprint: 8B8B 2209 0C6F 7C47 B1EA EE75 D6B7 2EB1 A7F1 DB43
"Everything should be made as simple as possible, but not simpler"
- Albert Einstein
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.7 (GNU/Linux)
iD8DBQE+H7c/1rcusafx20MRAjYNAKCW0lmjjzVnXo0CoAcOVj6lAV2jpQCglsjJ
mGjt9uV5BOCF9MwoE04SukE=
=mSOs
-----END PGP SIGNATURE-----
