I'm a windows user and i would like to make a mini cluster but i think windows wouldn't be as good as Linux in this application. i was wondering is i could use slitaz as the OS for the slave nodes and openSUSE as the master node, b/c i have that already installed on my main computer. if you could tell me if that would work please do but if you have anymore information about how exactly to make the cluster would be Greatly Appreciated. keep in mind im a Linux noob and should be talk to as one.. lol
I've been thinking about recompiling slitaz for x86_64 for this purpose. <Indirect solicitation for comments on how big a can of worms this is. >
The answer to your question depends on what you're doing and the iron you have.
If openSUSE qualifies on both counts then great, and consider it for compute nodes. Heterogeneous clusters can certainly be configured, it's just more of a pain that's better avoided absent some compelling reason.
Otherwise take a look at rocksclusters.org. Rocks is based on RHEL/CentOS/SL (CentOS default) but the current is 5.4 which is a little antiquated. It autoconfigures compute nodes and can automate reconfiguraton for any given compute task. It is a bit of an investment to learn, but it's high end. ScientificLinux is probably also a good choice. Slitaz is appealing in that it's compact and can run from memory so it can make older hardware reasonably useful (although once you get to 3 to 5 machines you have to look at whether your power mains can handle the load; FYI residential circuit breakers in the U.S. are usually rated at 2kW, although if you keep them near that load long enough they can supposedly weld themselves closed which is a fire hazard) but the trouble with using alot of older harware compared to a few newer machines is power (both delivery and cost) and heat.
I can't really comment about Windows clusters other than to express mild shock that anyone who knows what they're doing would do such a thing without some really really good reason.
Without knowing what you're doing it's hard to comment further other than to add that if you're compiling anything you should make sure that (assuming you don't already have some parallel compiler and were planning on using gcc) that the OS you chose has mpich or lam or something to handle interprocess communication if the code you're using doesn't take care of all of that itself, unless you're up to compiling these (intermediate to advanced, depending on luck) for the OS you have some other reason to prefer.
I'll add that SLES is used on a few high end clusters, so it's likely well adapted.
Xcat apparently supports memory image OS deployment to nodes, and purports to permit compressed images. I haven't tried this, yet. Note that by its developers' account, Xcat requires more effort to learn than Rocks.