As we moved from our older Exchange 2003 environment that was used primarily only for calendaring, to Exchange 2007 utilizing all of its functionality, we took time to architect the best possible environment. In our case, we have been utilizing server virtualization since 2005 and Hyper-V since the early Betas in 2008, so we knew from quite a lot of experience the capabilities of the virtual environment and wanted to utilize it as much as possible. In fact, to prove the design we built an entire testing infrastructure (12 VMs) within the Hyper-V environment to work out any problems and document the process. Without the virtual environment this process would not have been possible or as informative.
When we looked at the production environment and the necessary resource requirements of an Exchange infrastructure with around 8000 mailboxes we came to the conclusion that not all of the environment would be a good fit for virtual. The manageability of the disk space required for the mailbox servers, if they were virtual, was out of ‘our’ current mold for virtual candidates, so those were created as a pair of physical CCR clusters (4 total) to meet this need. But for ISA, Hub Transport, and CAS servers (2 each), these workloads work very well. This allowed us to avoid purchasing 6 physical servers and allowed for a design that had better redundancy since each of the VMs reside on Hyper-V clusters. Another benefit taken into account is that if we do see high resource utilization on any one of these workloads adding additional CPU, Memory or Disk resources is almost immediately available. Looking at the new dynamic memory feature coming to Hyper-V, these workloads may be good candidates.
For the full details check out the full case study here.