Good counterpoints. Now this is what we need on this thread, rather than twisters random blatherings. If all goes smoothly with either solution (and that means 1 hour of Sun downtime, 1 day of Dell downtime in your assumptions), then sure, you'd be crazy to purchase a $1M system that would be satisfied by a $400K system.
However, large systems scale in a nonlinear fashion, and who are you going to call when your server blue-screens. Microsoft systems will, and do, do that. Intel? Microsoft? Dell? Frys? Who. Or are you going to engage EDS to manage it for you, and who will they call. The point is, that there is no single entity with enough knowledge of the system from end-to-end to be able to track down random sporadic failures in unproven multi-processor systems. Intel don't have access to the Microsoft source code, and wouldn't know how to diagnose problems with it even if they did. Microsoft don't have engineers who could figure out hardware designs as complex as the latest Intel chips. And Dell don't have engineers who could do either. Compaq at least is trying: they acquired DEC in an attempt to build a systems business. But Compaq don't have the Microsoft source code or expertise, and their DEC purchase seems to have vaporized on them.
Who produces end-to-end systems right now: only IBM and SUN. The only other potential candidates are server-hosting companies like Critical Path and Exodus, but I'm not sure even they will have enough expertise to keep systems up and running in the presence of the inevitable failures. Also, interestingly they seem to be Sun houses. Perhaps they know something?
Note: this analysis does not mean that Solaris on X86 is a non starter. It would have to be in a system designed and built by Sun, and used internally by Sun, not just any-old-server-box with Solaris installed on it by an ambitious systems administrator.
P. |
