Installation guide

Chapter 3. Running a Windows Terminal Server and MetaFrame pilot 21
This graph shows that the average processor utilization was only 3%. This is well below the
average processor guideline of 80%. Based on the CPU results alone in this example, the
terminal server can support several hundred concurrent users.
You must also consider the application’s RAM requirements before a final number of users
per server. For the user’s RAM requirements, graph the per session Working Set values and
Available Memory to determine the average RAM requirements per user. Microsoft’s memory
guideline for WTS is to provide a minimum of 10 MB per user. As a general rule of thumb,
based on work in IBM’s performance laboratory, we recommend that you configure servers so
that average memory utilization
does not exceed 70%. Thirty percent is usually enough
extra memory so that the server does not expand storage onto disk or page memory onto
disk during periods of peak activity.
Using this memory recommendation, you can calculate the maximum number of terminal
server users, given a particular amount of server memory. Determine the maximum amount
of RAM your production servers will have on your virtual machines. You must use Windows
Server 2003, Enterprise Edition, if more than 4 GB of RAM will be installed.
Use the following formula to determine the maximum number of users from your pilot memory
results:
Maximum number of users per server (due to RAM limits) =
(Maximum server memory in MB - 128) x 0.7 / per user RAM required in MB
For example, if you determine from your pilot that the average per user RAM requirements is
35 MB and your servers will be installed with 4 GB of RAM, you can calculate the maximum
amount of users as:
Maximum amount of users = (4096 - 128) x 0.7 / 35 = 80 users
Based on your application’s RAM requirements, the number of users per server varies
depending on the applications installed, how much they are used, and how much memory
each application requires.
3.2.3 Interpreting your pilot bandwidth requirements
Bottlenecks in network communications can occur in four different areas:
򐂰 The client’s network interface
򐂰 The physical network media
򐂰 The server’s client-to-server network interface
򐂰 The server’s network interface for server to server/host communications
Bottlenecks in network communications directly impact the user at the client workstation.
When network delays are encountered, the delay in response time on the client workstations
is instantaneous. You may consider WTS performance unacceptable because of delays in
network communications even though the CPU and memory are available.
A server running Windows Terminal Server communicates with its clients via the RDP over
TCP/IP. Citrix MetaFrame communicates with thin clients by the ICA protocol. This protocol
can operate using transport protocols such as TCP/IP, NetBEUI, or IPX/SPX.
Both ICA and RDP are
on demand protocols. This means that when the display is static and
no keys presses, mouse movements, or clicks occur, no bandwidth is consumed. This
provides for much more efficient use of the network.
The amount of bandwidth that your Citrix users require can be determined by graphing the
per session Total Bytes as Kbps. Figure 3-2 shows the bandwidth measurements for ten