The challenge of setting performance thresholds
Setting meaningful and effective thresholds in performance monitors can be a challenge. You might try using standard rules-of-thumb or vendor recommendations, but the definition of normal behavior or good performance can vary, depending on:
- Site standards and practices
- The tasks being performed by the system, workload, or application
- The affected time period (time of day and day of the week)
To overcome these challenges, you could try a statistical approach--using your own data to determine what is normal for a particular resource and its associated performance metric. You could calculate historical average and standard deviation values for a metric, and define normal as a range of values centered around that average. However, this approach ignores variations in time and business activity by combining values for all time periods (some with high activity, others with low). Analyzing different time periods for dozens, if not hundreds, of metrics manually over different time periods would be complex and time-consuming. And subsequently defining threshold conditions and alarm settings for each metric would require significant additional effort.
The Threshold Advisor application and dynamic thresholds on the mainframe can help you address these challenges by:
- Collecting and analyzing large amounts of performance data
- Recommending threshold values based on the analyzed data
- Making those thresholds available across your BMC AMI Ops environment, in both product views and alarm definitions
- Applying those thresholds dynamically, according to time periods that match your business activity
Related topic