I'd have to disagree with you on the practicality of a cluster. We have 10 clusters here at work, each one being active passive supporting exchange with 7k subscribers per cluster. We are currently running at 99.999% availibility(202 dpm) for the year
.

That is exactly why you would want to cluster, availability. My point was it's hard to engineer and not practical in most cases. I use it for Exchange in an Active Passive configuration. The cluster shares the same quorum and one of the many controller cards corrupted it the quorum. With so many parts it was hard to find the actual hardware fault because it wasn't a black and white failure. So they can be much harder to troubleshoot than single boxes, plus the software alone will kill you with Microsoft. So it's not practical in most situations and not a whole lot of software supports it. The software has to be "cluster aware" meaning it takes advantage of the cluster APIs and knows the condition of the cluster so that it can initiate changes and redirect it's I/O to the active clustering components. My comment was fiscal and not technological


Now we are all talking about expensive advance clustering. NT has clustering services as well, so that is an option.

Heatwave is correct, to cluster you have to have redundant EVERYTHING, otherwise why bother? I even have redundant network paths.

This whole exercise seems like far more work and money in customization than it's worth. I'd just pick up a couple of barebones type boxes and build them up into a decent cluster. Less work, less hassle, less cash.
Sounds like allot of us have experience, so let us know what you plan because some situations won't work with clusters and I would hate to see anyone makes some of the errors I have.