White Papers

BP1013 Best Practices for Enhancing Microsoft Exchange Server 2010 Data Protection and Availability 23
Table 4 shows the Exchange KPIs that qualified the health of the mailbox database server during the
load of 5000 concurrent users while creating and maintaining the cumulative eight Smart Copies.
These values are the average of all the samples that were collected during the entire duration of the
simulation. We did not identify any specific spike on any of these counters even at the outset of the
snapshot process, therefore the identified KPIs where constantly under the recommended thresholds.
We omitted the values reported under lighter workload as they were less significant for the analysis as
they tracked even better performances.
Table 4 Exchange KPIs during cumulative snapshots with maximum workload
Performance Counters
2 active
nodes
1 active
node
Microsoft
Threshold
RPC information store KPI
MSExchange IS \ RPC Requests
1.3
5
< 70
MSExchange IS \ RPC Averaged Latency
1.7
1.7
< 10 ms
Active DBs KPI (databases and logs)
MSExchange Database \
I/O Database Reads (Attached) Average Latency
8.7 9 < 20 ms
MSExchange Database \
I/O Database Writes (Attached) Average Latency
2.5 2.9 < 20 ms
MSExchange Database \ Database Page Fault Stalls/sec
0
0
0
MSExchange Database \ I/O Log Writes Average Latency 1.8 2.4 < 10 ms
MSExchange Database \ Log Record Stalls/sec
0.01
0,02
< 10
Passive DBs KPI (databases and logs)
MSExchange Database \
I/O Database Reads (Recovery) Average Latency
8.2 N/A < 200 ms
MSExchange Database \
I/O Database Writes (Recovery) Average Latency
2.4 N/A < 200 ms
MSExchange Database \ I/O Log Reads Average Latency
2.5
N/A
< 200 ms
MSExchange Replication \ ReplayQueueLength (per database)
<=1
N/A
Low
We further analyzed the four pillar indicators of the system performances: disks, processor, memory,
and network, in the quest of a spike of any kind from the described course. All the explored trends did
not report any significant deviation, the only noticeable deflection from regular trends we could
identify was the one shown below in Figure 9.
While the average utilization of the disk/volume was not changed, we tracked a disk read spike lasting
a maximum of 30 seconds each time the Shadow Copy (freeze) initiated its activities. Figure 9 shows
the disk read spike on the volume and the disk write pattern which kept a steady trend instead. We also
show the disk read operations of the Exchange process (
store.exe ) as well, which we identified as the
source of the short spike.