The goal of this article is to give the reader an estimate of the relative impact that the Microsoft Lync 2010 desktop client has on the Microsoft Exchange 2010 Client Access server. This estimate can be used to determine the Exchange Server resources required to design a new or expand an existing Microsoft Lync Server 2010 deployment.

The Lync Server test team, in partnership with the Exchange Scalability and Capacity Planning team, investigated the capacity impact of running a Lync 2010 desktop client according to Lync Server 2010 User Models. The results of this testing, presented below, give you a better understanding of the impact that a Lync 2010 client may have on an existing or planned Exchange Server 2010 deployment.

Note. This analysis applies to Exchange Server 2010, Lync Server 2010, and the Lync 2010 desktop client.

Testing Environment

All tests were performed based on Lync Server 2010 User Models and were executed in an environment  similar to the environment  used in White Paper: Understanding the Relative Costs of Client Access Server Workloads in Exchange Server 2010 .

The test team used the topology and computer configurations described below to perform the tests.

Topology

The test topology included seven computers containing the following server roles or clients:

  • Three computers hosting the Microsoft Exchange Load Generator 2010 (LoadGen) client.
  • One computer hosting the Exchange Client Access server role, which hosts client protocols and services, such as Post Office Protocol version 3 (POP3), Internet Message Access Protocol 4 (IMAP4), Secure Hypertext Transfer Protocol (HTTPS), Outlook Anywhere, Availability service, and Autodiscover service.
  • One computer hosting the Exchange Mailbox server, which hosts user mailboxes and provides a single storage location for email and voice mail.
  • One computer hosting the Exchange Hub Transport server, which routes email messages from the Exchange Unified Messaging server to user mailboxes.
  • One computer hosting the domain controller.

Hardware Configurations

The Exchange Client Access server had two processor sockets on the motherboard populated with two Intel Xeon 5148 four-core processors running at 2.33 gigahertz (GHz), for a total of eight physical cores, along with 16 gigabyte (GB) of RAM. Hyper-threading was disabled on the platform to increase accuracy of the CPU cost computations. The Exchange Mailbox server used the same processor and RAM configuration as the Client Access server. Three mailbox databases were deployed on 3 separate spindles. All mailboxes were distributed evenly across the three databases. The Exchange 2010 Hub Transport server was configured with a single Intel Xeon 5148 four-core processor running at 2.33 GHz and 16 GB of RAM.

Testing Methodology

Tests were performed using the topology and computer configurations described in Testing Environment.

Each simulated Lync 2010 client ran for a period of three hours in isolation from other clients (such as Microsoft Outlook and Microsoft Outlook Web Access), against  Exchange Server  2010. Because only Lync clients were included in the testing, the measurements are an estimate of an actual load that may include additional clients or applications, such as Outlook Web Access, remote Windows PowerShell, POP, IMAP, and Microsoft Exchange ActiveSync.

In all, four tests were run using 1k, 3k, 6k, and 12k users, respectively. Throughout this document, the term “Lync 2010 user” refers to a single Lync 2010 desktop endpoint. For the purposes of this paper, if users commonly use multiple instances of the Lync desktop client concurrently, each concurrently running instance of the client counts as a “Lync 2010 user.”

Measurements

The following measurements were taken as part of each test run:

  • CPU usage on Client Access servers (CAS) and Mailbox servers  (MBX)
  • Latency (Average response time) from Exchange Web Service (EWS)

Thresholds Used

To ensure that there is enough capacity available for other applications, including Outlook Web Access, remote Windows PowerShell, POP, IMAP, and ActiveSync, on the Exchange Client Access server, the following thresholds were used to evaluate the recommended number of Lync 2010 desktop clients per Client Access server:

  • Client Access server total CPU should remain below ~10%.
  • Exchange Web Service process working set should grow steadily (100 MB in 6 hours).
  • Exchange Web Service average response time should remain below ~200 milliseconds.

These thresholds are recommended target values. Different thresholds may be more appropriate for your deployment configuration (for example, if more CPU is available for consumption of resources due to fewer applications being deployed or utilized, and so forth).

Results

The following charts summarize the results of the testing in terms of the CPU impact on the Exchange Client Access server.

CPU Trends

The following figures describe the CPU trends for the Client Access server (CAS) at the various testing points (1000, 3000, 6000, and 12000 concurrent users). For Lync 2010, the number of Client Access server cycles per user increases with additional users.

 

Figure 1. CPU trend for CAS CPU consumption per Lync 2010 User.

Applying a second order polynomial to this cost per user, we predict that approximately 27,000 Lync 2010 users would consume 75 percent of this server's CPU, as shown in the following graph:

Figure 2. Total CAS CPU Consumption per User

Average Response Times

The following diagram describes the average response times per request for the various test set sizes: 1k, 3k, 6k, and 12k Lync 2010 users.

Figure 3. Average Exchange Web Service response time per request.

Number of Requests

The following diagram describes the total number of Exchange Web Service requests generated across all users for the test set sizes: 1k, 3k, 6k, and 12k Lync users.

Figure 4. Total number of Exchange Web Service requests generated across all users.

Summary

With a 10% total CPU threshold and 200ms average Exchange Web Service request latency cut-off, each Client Access server supports an estimated 3,500 Lync 2010 users.

Table 2 uses numbers published in the Understanding the Relative Costs of Client Access server Workloads in Exchange Server 2010 white paper  to compare the performance of the Lync 2010 client to Outlook Web Access and Outlook.

Table 1. A comparison of client performance for different threshold values

CPU Threshold (%)

OWA Users

Lync 2010 Users

Outlook Users

75

14,000

24,000-27,000

34,000

50

9,333

19,000

22,666

10

1,866

3,500

4,533

The test environment used for the Outlook Web Access and Outlook numbers above is based on a 75% CPU threshold of 12 GHz. The 75% CPU threshold used for Lync is 14 GHz. This is due to a difference in the CPUs used in the two environments. In the Outlook/ Outlook Web Access environment, two Intel Xeon L5335 four-core processors running at 2.00 gigahertz (GHz) were used. In the Lync Server environment, two Intel Xeon 5148 four-core processors running at 2.33 gigahertz (GHz) were used. By comparing the MHz/user between the various workloads, it can be concluded that at the 75% CPU threshold on a similar processor, Lync Server would support approximately 24-27,000 users. At the lower end of the spectrum, the difference in CPUs is less predominant. 

The following table summarizes the raw data used for this analysis.

Table 2. Performance Metrics Summary

 

CAS

MBX

# of Users

% CPU (Total)

% CPU (w3wp)

Requests/sec

Average Latency

Working set (w3wp)

 

MHz/user (CAS)

 

% Processor time (Total)

 

IOPS/user (MBX)

 

1k

5.318

1.73

5.39

23.79

1.088 GB

0.0528

3.784

0.0018

3k

8.766

4.73

14.35

24.318

1.365 GB

0.0290

4.748

0.001

6k

15.144

10.33

28.812

25.747

1.512 GB

0.0251

6.542

0.003

12k

29.088

22.61

57.496

49.713

1.439 GB

0.0241

9.401

0.0058

Lync Server Resources

We Want to Hear from You

 

 Keywords: Lync 2010 desktop client, Exchange 2010 Client Access server