This is something which I’m really pleased to see Microsoft talking about, whilst technically supported using NLB for CAS it’s a bad solution, some of the reasons being:
- It doesn’t scale well
- It isn’t service aware (if the NLB service is running on the target server it will get its share of client requests regardless of if Exchange is actually able to service them)
- You cannot combine it with Failover clustering (which is required for a DAG node) & multi role solutions (Hub, CAS & Mailbox) are very effective with Exchange 2010
- Potential switch issues
So as Steve Goodman has reported Ross Smith IV spoke about this in his TechEd Europe session a few weeks ago his recommendation is to use a proper HLB solution, I’ve had good results with the Kemp series of HLB’s, I’ve used both the virtual appliances and the physical hardware appliances. These are very cost effective & meet the requirements for load balancing & making Exchange 2010 CAS highly available. Henrik Walther has written a good article on using the Kemp appliances with Exchange 2010, well worth a read.