Global performance considerations

There are four main performance considerations that apply to all types of Exchange Servers running Forefront Protection 2010 for Exchange Server (FPE):

Engine management

Scanning navigation

Engines and performance

Process counts

Engine management

Intelligent Engine Management (IEM) automatically selects the engines used in each scanning process on a particular server. This is typically the available five engines. You can also manually select the number of engines that should be used in each scanning process for protection. This decision impacts throughput performance and memory consumption.

The base memory footprint for each scanning process is 350 megabytes (MB) for the default five engines used within each scanning process. This number can be reduced by approximately 63 MB per engine if you reduce the number of engines being used. For a scanning process with only three engines, the memory footprint of the scanning process is approximately 224 MB. Note that this calculation must be multiplied by the number of scanning processes on a server where the default for a non-multi-role server is four scanning processes. In addition, each server requires an additional 650 MB of memory to support the underlying services. The following table depicts the memory requirements for standalone Hub and Edge servers with varying numbers of cores, engines, and scanning processes.

For more information on calculating the requirements, refer to the FPE Capacity Planning Tool.

Scanning navigation

The performance of Exchange Server 2010 is correlated to the size of the actual messages being processed. FPE is not only dependent upon file size but also on actual file types. Zip files and other container files introduce a larger overhead than simple processing of .gif or .jpeg attachments. For example, the Microsoft Word 2010 format, OpenXML, is actually a nested .zip file.

It is challenging for an organization to determine how these variables will impact their overall performance because different business verticals have different size and attachment characteristics. The data provided in the capacity planning tool uses a representative message set. Data is collected from several different business verticals. The data is then evaluated based on file type complexity and size. The representative message mix is created to be a near worst case combination that can be used conservatively to evaluate the performance of a system regardless of business vertical.

Engines and performance

In the FPE capacity planning tool you can select from among four options under the Engine and Performance category:

• Scan with all engines

• Scan with the subset of engines that are available

• Scan with a dynamically-chosen subset of engines

• Scan with only one engine

For information about these options, see Configuring the number of scan engines used for each scan. The following graph shows the performance of an FPE Edge server with various numbers of engines and Engines and Performance selections. Note that the message rate depicted in the graph denotes the maximum message rate before queuing starts. For example, a 5 Engines/Subset 4 core server will start queuing at a load greater than 22 messages per second.

Process counts

By default, there are four processes associated with an FPE server role. However, you can reconfigure the number of processes for the transport, realtime, and scheduled scans via their respective Process count settings (for more information, see Configuring the transport scan, Configuring the realtime scan, and Configuring the scheduled scan). The following graph shows the difference in overall throughput and capacity for an eight core computer with four and eight processes. Note that neither of the configurations burdens the CPU of the server; however, the maximum throughput gained is approximately 30 messages per second for a four process configuration versus approximately 54 messages per second for an eight process configuration. This performance data is achieved without queuing. For transport and realtime scanning, it is recommended that you set the number of processes on individual FPE servers to the number of cores on the targeted hardware.