Troubleshooting

Migrating Exchange 2010 To Dell Advanced Infrastructure Manager Environment
Page 23
Disks (per storage unit)
16 x 500GB 7.2k SATA
RAID
2 x { RAID 10 of 14 disks + 2 hot spares }
LUN‘s (per storage unit)
2 (1 active / 1 passive)
Total Number of Active Databases
2
Total Number of Passive Databases
2
EqualLogic Controller Firmware
5.0.5
Number of pools
2 (1 member per pool)
Memory Requirement = 4 GB + 9MB per mailbox x 1800 users = at least 20GB.
Settings for Jumbo Maximum Transmission Unit (MTU) and flow-control were enabled on the switches
and server ports used for iSCSI communication. Each M610 had an active and a passive database copy
that were stored on a PS6000E array using 500GB SATA disks in RAID 10. The results collected are
summarized in the succeeding sections for Exchange 2010 stand-alone and Exchange 2010 with AIM.
The deployment with AIM consists of performance measurements for persona in access mode and trunk
mode scenarios. In all cases, the results are well within thresholds proposed by Microsoft. Three
scenarios were tested and validated. Loadgen tests were run for all three scenarios, and results are
expressed as the average of results from multiple tests.
Scenario 1: Standalone Exchange 2010 ecosystem, without AIM‘s management. Results pertaining to
this scenario are considered baseline results.
Scenario 2: Exchange 2010 integrated within AIM environment and Exchange servers are AIM-managed
with personas in access mode. Jumbo frames and Flow control were enabled during this scenario. This
configuration does allow Exchange to fail-over from one server to another within the AIM pool, but does
not offer NIC redundancy.
Scenario 3: Exchange 2010 integrated within AIM environment and Exchange server is AIM-managed
with personas in trunk mode. Personas in trunk mode allow not only failover within servers in a pool,
but also offer NIC failover within a server.
Database Latencies
The most important consideration in validating the Exchange deployment is whether or not active
database latencies are below their targets. Active database read and write latencies must be below 20
milliseconds (ms). Passive database read and write latencies can be much higher, and on average are
less than 200 ms. Latencies were measured using the MS Exchange Database counters. The read latency
is the time taken to read from a database file, and the write latency is the time taken to write to the
file. The results indicate no performance degradation from Scenario 1 to Scenario 2 or Scenario 3.
Figure 8 shows the active database latencies and Figure 9 shows passive database latencies for all the
three scenarios. The latencies associated with active database copies in non-AIM as well as AIM
environment are well within the threshold. With Exchange servers in AIM environment, there is
increase in latencies associated with passive database copies.