Overview of Performance Monitoring

Controlling Performance Monitoring Overhead

When you select a counter in any view, the performance tools collect data for all counters of that object, but display only the one you select. This causes only minimal overhead, because most of the tools' overhead results from the display. You can control monitoring overhead in the following ways:

Use logs instead of displaying a graph. The user interface is more costly in terms of performance.
Limit the use of costly counters; this increases monitoring overhead. For information about costly counters, see the Performance Counter Reference on the Microsoft® Windows® 2000 Resource Kit companion CD.
Lengthen collection intervals if possible. In general, 600-second (10-minute) intervals are sufficient for ordinary monitoring.
Collect data during peak activity rather than over an extended interval.
Reduce the number of objects monitored unless these are critical to your analysis.
Put the log file on a disk that you are not monitoring.
Check the log file size when logging multiple servers to a single computer to see how much space the data is taking up.
Limit to brief periods the trace logs that are monitoring page faults or file I/O. Prolonged trace logging strains system performance.
Avoid configuring System Monitor reports to display nondefault data. If you choose nondefault data (the defaults are Average value for logs; Last value for graphs) in the Report view, the statistic is calculated at each sample interval. This incurs some additional performance overhead.