White Papers

BP1026 Sizing MS Exchange with EqualLogicPS6100 & PS4100 on VMware vSphere 5 12
Exchange messaging service end users retrieve and store their data on the server databases traversing
the Exchange Server services. In turn, the Exchange Information Store service running on the mailbox
server benefits from the database cache to perform its own storage access instructions via the ESE
interface. The database cache is retained in memory, thus the access to and from it is considerably
faster than if performed with the storage subsystem directly. Furthermore, Exchange utilizes internal
algorithms to combine changes to the same blocks of data before flushing them to the disks,
achieving an even lower amount of IO access.
Formerly, the metric adopted for Exchange user activities was founded on light to heavy classification
with intermediate degrees of workload. Currently, the user profiles are identified by the association of
the number of messages sent and received per mailbox per day, where the average message size is
75KB, and the corresponding amount of memory allocated in the database cache per each mailbox.
The three profiles selected for the test, reported in the Table 2, specify an average mailbox workload
progressing from 50 messages to 150 and then to 300 messages per day. The corresponding average
cache allocations for the same mailbox are estimated to be 3 MB, 6 MB, and 18 MB. The final
estimated IOPS per mailbox, according to these profiles is evaluated in 0.06, 0.18, and 0.36 IOPS.
When applying this count to our simulated scenario of 5,000 mailboxes, the industry accepted average
user count for a medium enterprise, hosted by a single server, we account a total estimated IOPS
progression (transactional only) of 300, 900, and 1,800 IOPS. These estimated IOPS values do not rate
all the remaining Exchange Server activities generating IO access. The database cache evaluated for
this pool of users would be 15 GB, 30 GB, and 90 GB; again without considering the additional
memory requirements due to other factors.
The results collected from the Exchange Jetstress simulation of these three workload profiles are
shown in Figure 4.