Okay, so I'm well on my way to building a Beowulf Cluster (yes, Kenny, turns out that's what I was building all along) for my project with Philip on neural nets and genetic algorithms.

We've decided on the basic architecture and software:

Master Node (Micron PC):
1GHz/512MB RAM
Hard Drive 1 (10GB) -- Windows XP
Hard Drive 2 (6GB) -- RedHat Linux 9.0

Slave Node (Compaq AP400 Workstation, 2 so far):
PII 400MHz x 2
256MB RAM
RedHat 9.0

The specs are slightly dated, but the price is right. There's a local vendor that can give me nearly all the Compaq AP400s I want for $65 a piece, and the cost of upgrading them to the specs above is about $35 per node, which means that a 10-node cluster, with 20 processors, will cost me about $1000.

Also, the AP400s are upgradable to dual PIII 700 MHz/1GB RAM, so they're fairly scalable if I want to dish out more money.

This seems like a pretty good cost/performance ratio, but I'm admittedly a neophyte.

We're programming in pure Java, and we'll be using JCluster as our distributed computing environment.

I should have an 11-15 node cluster set up and ready to crunch numbers within a couple of months.