Folks,
Our company has 3 Suse Linux Servers and I've been thinking of throwing out there the idea of clustering all 3 servers together, because of all issues we've been having with running processes in separate servers. Does Novell offer a cluster solution for their Enterprise Servers? If not, is there anything out there that would allow me to do implementation?

Thoughts?

Before you throw it out there, have you considered what tasks you would use a cluster for? Do you run a lot of serial jobs that you can queue up? Do you run lots of parallel jobs that could use MPI?

Also, you say you have 3 servers. How many processors/cores are you talking about? Is this a case of 3 physical processors each with 4 cores? (thus your effective cluster size is 12 cpus) or are their more/less cpus? Generally speaking, the more cpus you have, the "better" a cluster. Mind you my last statement is a _very_ big generalization, a cluster is only as good as the use you can make of it.

I've used ROCKS cluster software at my work and I've been pretty happy with it.

HTH

Gattaca,

Thanks for your reply, yes, as I said, we currently have 3 Linux server with 4 processors and about 4GB of memory each. These servers are used for ETL (Extract Transform and Load, Data) processes and since our data universe is growing constantly, these servers are becoming relatively small for these processes, I was thinking that if we cluster the 3 servers all together, we could make a better used of both, processors and space.

Yes, we do a lot of serial jobs, we never used any type of technology that would allow us to queue these jobs up, not even sure how to do that, even though I don't believe this would be a great idea, since we live in a very tight schedule. The real solution would buy another server, but we don't have money on the budget for that right now, so I thought that another solution would be clustering the servers.

I'm not sure what MPI is, please explain.

Correct, 4 CPUS/4 Cores.

So, should I "cluster"?

First some clarification: In total you have 12 processors? --> 3 computers with 4 processors each = 12 total processors right?

The whole point of a serial-job cluster is to distribute jobs sequentially to processors as they become available. In a perfect world, the user only has to submit one big job to the cluster and then gets an email when all of the work is done.

If you only run serial jobs but you don't think queuing up jobs to be done one after another is a good idea, then clustering is not for you.

First some clarification: In total you have 12 processors? --> 3 computers with 4 processors each = 12 total processors right?

Correct.

The whole point of a serial-job cluster is to distribute jobs sequentially to processors as they become available. In a perfect world, the user only has to submit one big job to the cluster and then gets an email when all of the work is done.

I misunderstood then.

What did you misunderstand? The purpose of a cluster? or what I wrote? I'm not the clearest writer so I just want to make sure I haven't put you off clustering just because I can't write.