Performance monitoring best practice
Updated: January 21, 2005
Best practices
-
Set up a monitoring configuration. Configure Performance Logs and Alerts to report data for the recommended counters at regular intervals, such as every 10 to 15 minutes. Retain logs over extended periods of time, store data in a database, and query the data to report on and analyze the data as needed for overall performance assessment, trend analysis, and capacity planning.
For best results, do the following before starting System Monitor or Performance Logs and Alerts on the computer you want to monitor for diagnostic purposes:
- If the server in question has halted or is not responding, run System Monitor from another computer.
-
Keep monitoring overhead low. In general, the performance tools are designed for minimal overhead. However, you might find that the overhead increases under the following conditions:
- You are running System Monitor in graph view.
- You have selected an option other than the default (current value) for either the System Monitor graph or report views.
- You are sampling at very frequent intervals (less than three seconds apart).
- You have selected many different objects and counters.
- Other aspects of performance tool operation that affect
performance include file size and disk space taken up by log
files. To reduce file size and related disk space usage, extend
the update interval. Also, log to a disk other than the one you
are monitoring. Frequent logging also adds demand on disk input
and output (I/O).
If monitoring overhead is a concern, run only the Performance Logs and Alerts service and do not monitor using a System Monitor graph.
During local logging of remote counters, frequent updating can slow performance due to network transport. In this case, it is recommended that you log continuously on remote computers but upload logs infrequently -- for example, once a day.
- Analyze performance results, and establish a performance baseline. Review logged data by graphing it using the System Monitor display or exporting it for printing. Compare the values against the counter thresholds shown in Analyzing performance data to verify that resource usage or other activity is within acceptable limits. Set your baseline according to the level of performance that you consider satisfactory for your typical workload.
- Set alerts. Set alerts according to the counter values you consider to be unacceptable, as defined by baseline evaluation.
- Tune performance. Tune system settings and workload to improve performance and repeat monitoring to examine tuning results. For more information on ways to improve performance, see Solving performance problems .
- Plan ahead. Monitor trends for capacity planning and add or upgrade components as needed. Maintain logged data in a database and observe changes to identify changes in resource requirements. After you observe changes in activity or resource demand, you can identify areas that might require additional resources.