At times it is not possible to run performance
testing benchmarks under identical conditions. For example, the load
may be slightly different, or the measurement duration may differ,
or you may want to compare the same activity at different times of
the day or week. To compensate for these differences, you can scale
one set of data to more closely match the other
set.
Scale and Scale Special are available from the Edit menu on the profiling viewer when comparing two profiling data files.
Use these scale options when the data files that you are comparing
are unequal in some way.
For example, imagine a scenario where you are
able to generate a constant load on your systems, but one profile
represents 10 minutes of application activity and the other only 5
minutes. You can use Edit
Scale Special to set and apply a factor
of 2 to the shorter time period, which normalizes the 5–minute
data set to 10 minutes for easier presentation and interpretation.
Or, suppose that you want to look at activity
for a particular application during peak business hours and compare
it to a relatively quiet period at night. You know that the number
of transactions will differ, so how do you compare the amount of work
done for each transaction?
Open a profiling viewer with the peak time data.
Open another viewer with the night data. In the peak time viewer,
click File
Compare to open a profiling viewer with the comparison data.
In the comparison viewer, select Metrics
Code/CPU
Method Call Count. Click to select a known
method that represents a unit of work or a transaction. Then click Edit
Scale from the menu bar. HPjmeter automatically recalculates the call
count for each method in the peak data with the result that each value
is normalized to the ratio of the call counts for the selected method.
HPjmeter then sorts the data. By viewing the information
that appears in the newly sorted list, you can compare differences
in activity between the two data files. If you want to see how this
works in detail, select View
Show Formula to display the calculations
that HPjmeter uses.