This is exactly what I was thinking. Start looking at IOPS. Determine the IOPS characteristics of your users and then build disk subsystem to accommodate it. You generally have 3 levels of users and a percentage at each level. Also make sure you have enough GCs in your environment. Sounds like there should probably be a lot more homework before any decisions are made. Make sure you watch the exchange web casts available. Don't think they are for dummies. A couple of them have some really top notch information in them. _____ And what is the calculated IOPS per user? Basing on what I know about other large deployments, I would guess that you are probably OK assuming your disk subsystem can meet the IOPS load. In the general case, as I'm sure you know, Exchange is I/O bound, not processor bound. Are you going to have a dedicated bridgehead server plus a couple of FE servers? Seems like that would be a good idea too in this size of deployment.