I’m a huge believer in all things virtual. I’ve even written articles stating that I believe that workloads should be virtualized whenever possible and reasonable. However, that doesn’t mean that one should blindly virtualize just to virtualize. In fact, there remain workloads and situations in which it’s perfectly acceptable to deploy physical servers. I’m working on one such project right now.
I’m working with a client on a 9,500 seat Exchange 2010 deployment. This is a relatively unique deployment with a ton of mailboxes but only a fraction of them used simultaneously and many used only sporadically. The client also wishes to implement database availability groups (DAGs) for mailbox failover and redundancy.
The client has a significant VMware vSphere environment that runs most of their systems. The vSphere hosts are connected to a NetApp storage array.
I’m recommending to the client the following:
- Host the Exchange Client Access role on multiple virtual machines and create a CAS array. Use VMware DRS affinity rules to prevent CAS servers from ending up on the same host.
- Host the mailbox and Hub Transport server roles on physical hardware with direct attached storage
- The NetApp space is very expensive. Use direct attached storage to keep the solution cost reasonable and to avoid overtaxing the NetApp.
At present, the organization won’t be deploying the Unified Messaging or Edge Transport roles, so they’re skipped.