I'm designing a multi-machined system that is supposed to resemble a failover redundant web&&databaseserver. I think I have the theory pretty well figured out (which means I'll run into a lot of unexpected problems )... but my concern is this:

I need some extra hardware, and I can buy three PII's for almost no money, or a couple of PIII's that cost apiece what the PII's cost in total. Is there a good reason to choose one over the other?


One day in the distant future the setup will go live on x64 hardware. So I'm hoping to transport as much of the configuration of the test setup to the new setup. Would that be a concern somehow?

EDIT:
Maybe I should explain, it does not really have to perform in any sense, but prove workload gets evenly distributed between two (or more) machines and that it can handle it if the plug gets pulled from a machine.

If I understand correctly, you will be buying this hardware just for tests, and not for "the final product".
In this case, I would suggest doing your tests in a VM as it's free

There are a lot of virtual machine apps out there so you'll have check which one suits you best.

My personal favorite is the KVM + VDE combo

Thanks,

There are subtle differences between networked VM's and iron && copper networks. I've gone wrong before because I assumed it would all be the same.

In the final setup, I want to get the different tasks the system needs to attend to as compartmentalized as possible. That will also mean running VM's. Things like webservers for example will not run on iron in this case. But that's all at a later stage.

First step, and that to me is complex enough, is to be able to pull the plug from a machine and watch the others compensate. Then the plug back in, and then watch all adjust accordingly. And then joy and (perhaps) profit.