>Again, the idea is to use one machine per COM process, as my analysis indicates it's cheaper to load up on machines than anything else.
>
>Well, Ed, how about this: What would you do?

I don't recommend Win9x as a platform for servers. Period.

If you're primary concern is cost per system, don't bother asking. NT costs more than Win98. You're going to go out and buy Win98 boxes and kludge up a job server farm. Don't let me stop you. It's not a server platform in many respects, but none of them is significant.

Let me ask a question. How do you intend to balance the load of jobs between servers, and how do you intend to ensure that only one remote server instance is going to be run per station? Without some sort of resource pooling, this isn't possible; neither is ubiquitous addressing of servers or blocking new instantiation of servers.

Perhaps COM (in this case, specifically DCOM is a bad implementation here. Your problem description requires asynchronous execution of requests, ubiquitous server addressing and control of server instancing. Without a server pooling mechanism, this is not trivial in VFP. I can think of plenty of non-COM schemes that might address this better.

If initial cost of purchase is the issue, I have nothing to say that will change your mind. If system capability, anonymity and reliability is the issue, it's a bad choice, but it may be better than the alternative of not doing anything. Your call here.