White Papers

A Sizing Study of Lync® Server 2010 and its Back End SQL Database on Dell™ PowerEdge™ Servers
21
instances when the SQL Server has to fetch a page from disk after it was not found in its memory. The
page faults/sec counter measures the soft page faults in addition to the hard page faults. These soft
page faults happen when pages are found at other locations within main memory and these are not as
expensive as retrieving pages from disk. Finally, the page life expectancy of less than 3600 (or 1 hour)
indicates that the server has run out of memory with progressively lower values indicating more critical
memory issues. Hence, it is desirable to have values greater than 3600 for these counters.
For Logical Disks, the reads/sec, writes/sec and transfers/sec counters have to be looked at taking in
to consideration their corresponding latencies. A latency value higher than 10ms is acceptable but
latencies greater than 20ms indicate an IO bottleneck. In the lab deployment, these logical disk
counters were measured for the Back End SQL databases and logs contained in LUNs 1, 2, 5, and 6,
each containing database or log files as explained in Table 3. The disk queue length is recommended to
be below 24 as suggested by Microsoft.
The MSSQL Buffer Manager counter for cache hits indicates the number of page reads or writes issued
to physical storage, and because this is expensive, it is recommended to be well above 90%. This will
ensure that the server does not have to issue requests for pages from disks and can instead use its
memory contents.
Load Generation Performance Counters
In addition to the counters from the Lync VMs and their host machine, counters from the load
generators were also collected to verify that the load generation system did not introduce latencies.
The major performance counters are in the Load Generation Counters Table below.
Load Generation Counters Table 9.
Performance Counter
Threshold
CPU Utilization
<60%
Available Memory
>15%
Network utilization
<50%
On the client machines, the CPU, memory, and network utilization were set below the acceptable
limits of the Hyper-V Host, so that the desired load can be generated for the Lync Servers. Ensuring the
CPU and available memory counters are within these thresholds eliminates the possibility of a
bottleneck at the load generator machines. To verify that the Lync system was healthy, the following
client counters were also recorded:
 Total Active Endpoints
 Presence Pass Rate Percentage
 Total IM Calls Active
 Total Number of IM Conferences Active
 Total Number of AV Conferences Active
 Total Number of AS Conferences Active