This blog post was written by Todd Muirhead, VMware Staff Engineer from the Performance R&D team.

I worked with the Dell TechCenter last year on some testing with Exchange 2007 on PowerEdge M710 blade servers and EqualLogic iSCSI storage. The Exchange environment was all virtualized on VMware vSphere and we tested to show how load balancing with DRS provides performance improvements. When Exchange 2010 was released at the end of last year, it was time to do some new tests.

Around the beginning of the year I acquired a brand new Dell PowerEdge R710 with two quad-core Intel Xeon X5570 processors and 96GB of RAM. The reason for the new server was to be able to test the new Microsoft Exchange Server 2010 running on VMware vSphere in virtual machines. Combining this shiny new server with a Clariion CX4-960 fibre channel storage array made for a great Exchange 2010 VM test bed.

The results have been impressive with Exchange 2010 running extremely well which made it easy to put together a three part series of blog posts (all links can be found on Dell TechCenter).

The first blog focused on the performance of a single Exchange 2010 Mailbox Server VM as it is scaled-up to support more users. The tests ranged up to 8000 users in a single VM. Performance remained good with low CPU utilization and fast response times.

Part two of the series looked at how adding more RAM to an Exchange 2010 Mailbox Server VM results in fewer IOPS. Performance tests showed that increasing RAM resulted in fewer IOPS and improved response time. It is possible make the tradeoff of additional RAM for less IOPS or fewer disks.

Rounding out the series was a look at scale-out performance. Using up to four Mailbox Server VMs to host up to 8000 users the tests showed that response times increased only slightly when running the same number of users across multiple VMs as when running them on a single larger VM. The reason for this slightly higher response time is because of a smaller percentage of RAM being available for use by Exchange to cache resulting in more IOPS.

Along the way I also published a few blog articles on my VMware community blog with some additional performance details that didn’t make the main series of articles. This included an article on the CPU utilization of the different Exchange Server roles when broken out into separate VMs, another on CPU usage for the ESX host, and finally a look at the I/O sizes for Exchange 2010 as measured.

A key takeaway from all of this performance testing with Exchange 2010 was that the Dell PowerEdge R710 was more than capable to host a number of Exchange 2010 VMs running a variety of roles and configured in a variety of ways. All the performance testing done showed that 95^th percentile SendMail latency times never exceeded more than half of the 1 second threshold at which users can detect a slowdown in performance.