[MEngineer]

I am a student on a university campus. Because of this, I have access to some of the "old" computers that are being phased out of the computer labs on campus. On that note, I have access to multiple computers of the exact same specifications. I would like to, in some way couple them together so that the processors, motherboards, and graphics can all work in unison. They will be running XP Enterprise, but I have access to any of the windows OS up to Win 7 through our program.

Usage we are looking to use this aforementioned machine to work with a server tray (also received through the re-use program) to do some network computing simulations, crunch some gnarly code, and just be all around awesome. We might consider attempting the same concept (if it is feasible) with a pair of PowerMac G5's...

I hope you geniuses can help me out.

 

[emosun]

It's funny because this was my very first question I asked this forum when I first joined. The basic answer is the machines cannot be physically joined together to make one faster pc , but rather through a network as separate machines working toward completing a task like folding or rendering or calculations.

 

[LMiller7]

What you are considering is known as distributed computing and is a viable concept. However, this is useful only for some very specific types of applications. Setting this up is not a simple thing, far beyond what could be described in a forum post.

 

[Vadigor]

As LMiller7 said this is possible in the form of distributed or parallel computing, which is used extensively in universities to construct inexpensive supercomputers. What you are describing would be called a Beowulf cluster.

However such a cluster only lends itself to tasks that can be run parallel, in multiple threads. There is a reason you'll only find these clusters in universities or specialised industries as only they generate the right type and amount of work for such a cluster to function at optimal capacity.

In addition to this, you'll require a moderate to extensive grasp of *nix, scripting, networking and systems engineering to configure such a cluster. Furthermore, the computers you describe would be suboptimal for this configuration as they are likely too large and consume too much power for the performance/cost metric to be positive. You'll also want at least half a dozen nodes in your cluster before you start getting noticeable performance.

Should you wish to go ahead, research OpenMPI and PVM. Both also work on Windows but I'm not sure how their performance is. Linux is the better bet as there are more light-weight versions of it available.

An alternative use would be to set up a render farm but again the same concerns apply. DrQueue is one example of a render farm that you should be able to configure with minimal prior experience. Odds are that your systems are too weak to be worth using.