Determining the Cause of Performance Problems
Analyzing Performance Results
261
2
Create a Performance report using only the interval specified by these two time
stamps.
3
Graph the Performance report to verify that the distribution has flattened.
4
Run the Performance, Compare Performance, and Command Usage reports to
examine the summary statistics for this measurement interval.
Determining the Cause of Performance Problems
The third level of analysis helps you understand the causes and significance of
performance problems.
Analyzing Results Statistically
This detailed analysis takes the low-level data and uses statistical testing to help draw
useful conclusions. Although this analysis provides objective and quantitative
criteria, it is more time consuming than first- and second-level analysis and requires a
basic understanding of statistics.
When you analyze your data at this level, you use the concept of statistical significance
to help discern whether differences in response time are real or are due to some
random event associated with the test data collection. On a fundamental level,
randomness is associated with any event. Statistical testing determines whether there
is a systematic difference that cannot be explained by random events. If the difference
was not caused by randomness, the difference is statistically significant.
To perform a third-level analysis, run the Performance and Response vs. Time reports.
Some of the measurements to consider during third-level analysis are:
s
Minimum
The lowest response time.
s
Maximum
The highest response time.
s
Mean
The average response time. This average is computed by adding all of the
response time values together and then dividing that total by the number of
response time values.
s
Median
The midpoint of the data. Half of the response time values are less than
this point and half of them are greater than this point.
s
Standard Deviation
How tightly the data is grouped around the mean.
s
Percentiles
The percentages of response times above or below a certain point.
The 90th percentile is often measured.
s
Outlier
A value that is much higher or lower than the others in the data.
