Virtualizing Microsoft Exchange enables us to host tens of thousands of Exchange 2013 mailboxes on a single physical server. In the blog post Exchange 2013 on Hyper-V, I discussed how we were able to host 66,700 Exchange 2013 mailboxes on a single physical server, a record! The storage used in that test was a NetApp FAS storage system (a two-node cluster) running Clustered Data ONTAP 8.2.
One of the capabilities of cDOT is scale-out. So, a natural question to ask is this: by doubling the number of servers and the number of nodes in the cDOT cluster, can you double the number of Exchange 2013 mailboxes, from 66,700 to 133,400?
The answer is yes. Figure 1 shows the linear scaling of the Exchange 2013 mailboxes with the number of cDOT cluster nodes.
Figure 1. Scaling Exchange 2013 mailboxes with cDOT nodes.
That’s great. But, just doubling the number of mailboxes is not enough. Can you double the number of IOPS and keep the latencies the same? The answers are yes and yes. Figure 2 below shows that we not only doubled the number of IOPS, but also kept the average read and average write latencies at the same level.
Figure 2. The Jetstress test results on 16 VMs. MBX11 to MBX18 were hosted on server1, and MBX21 to MBX28 on server2.
Figure 2 summarizes the test results on the per-VM basis; and 16 VMs were used to run the Jetstress test. The achieved IOPS exceeds the target IOPS by 43%. The average database read latency is 10.27ms and, write latency, 1.65ms. Both are far below the 20ms threshold.