The Cost of Performance Monitoring

Performance monitoring tools are quite sophisticated, but they are plagued by the problem common to all experimental tools: Using them changes their results. Performance tools are just applications and, as such, they occupy the processor, use memory and disk space, and tax the graphics subsystem of the Windows NT Executive. Make sure to measure the effects of these tools, and subtract them from your data.

Note

Performance Monitor for Windows NT 4.0 has lower overhead than previous versions, due almost entirely to changes in the Windows NT 4.0 architecture. Most of Performance Monitor overhead is consumed by its graphic displays, which are now more efficient, not by data collection.

Response Probe, a monitoring tool included on the CD, has no apparent overhead. It monitors its own toll on the system and subtracts it before displaying its results.

• To monitor Performance Monitor, include the Perfmon.exe process and its threads in your logs and charts and subtract them from your data.

• To determine how much disk space is consumed by Performance Monitor, log updates by doing a series of manual updates and watching the change in the log file size on the status bar in Log view.

• To measure the cost of monitoring particular objects, record the change in file size while adding and deleting those objects from a chart.

• If the cost of monitoring is too high, lengthen the Performance Monitor Update Interval to at least 15 seconds. Change the Task Manager Update Speed to Low.

• After you have collected baseline logging data, use Alerts to warn you of discrepancies. Alerts have the least overhead of any Performance Monitoring method. You can also log data over the network if you're studying disk performance.