Monitoring and Tuning Your Server

Analyzing Processor Activity Data

Of the counters listed in Table 5.4, the System\ Processor Queue Length counter is probably the most important for analyzing processor activity data. This counter displays the number of threads waiting to be processed in the single queue shared by all processors. Sustained high rates of processor activity, which leave little excess capacity to handle peak loads, are often associated with processor bottlenecks. Processor activity itself only indicates that the resource is being used, not that maximum use of the resource is a problem. However, a long, sustained queue indicates that threads are being kept waiting because a processor cannot handle the load assigned to it.

A sustained processor queue length of two or more threads (as indicated by the System\ Processor Queue Length Counter) typically indicates a processor bottleneck. You can set a PerfMon alert to notify administrators when the processor queue length reaches an unacceptable value.

The Processor\ % Processor Time counter is most often used as a general measure of processor activity on both single-processor and multiprocessor computers. System\ % Total Processor Time is included for monitoring system-wide processor use on multiprocessor computers. On single-processor computers, System\ % Total Processor Time always equals Processor\ % Processor Time. On multiprocessor computers, System\ % Total Processor Time represents the active time of all processors divided by the number of processors.

If the server workload is shared equally among all processors, System\ % Total Processor time is an excellent measure of processor activity. However, this counter hides bottlenecks resulting from unequal processor loads. (If one processor is 100 percent busy and three other processors are idle, the % Total Processor Time is 25 percent.)

Windows 2000 Server is designed for efficient scaling and includes several strategies for balancing processor load. An application, however, can create an imbalance by setting a processor affinity, which binds a process to a single processor. For detailed processor monitoring, you need to chart Processor\ % Processor Time for each processor on the computer.

It’s not unusual to encounter the following challenges in analyzing processor data:

A large processor queue when all processors are busy. Create a histogram of Process\ % Processor Time for each process. To do this, click the View Histogram button on the toolbar in PerfMon. The histogram shows the processor time consumed by each process.
A single bar in the histogram rises above all of the others. The process represented by the bar may be consuming a disproportionate share of processor time and causing a bottleneck. Consider replacing the application running in that process, or moving the process to another server.
The processors are being shared equally by several processes. Consider upgrading or adding processors. (Multithreaded processes benefit most from additional processors.)

For more information about processor use by applications related to IIS, see Web Applications.