HP MC68360 User Manual page 260

Chapter 8: Making Software Performance Measurements
Activity Performance Measurements
Error Tolerance and Confidence Level An approximate error may exist in
displayed information. Error tolerance for a level of confidence is calculated using
the mean of the standard deviations and the mean of the means. Error tolerance
gives an indication of the stability of the information. For example, if the error is
5% for a confidence level of 95%, then you can be 95% confident that the
information has an error of 5% or less.
The Student's "T" distribution is used in these calculations because it improves the
accuracy for small samples. As the size of the sample increases, the Student's "T"
distribution approaches the normal distribution.
The following equation is used to calculate error tolerance:
Where:
O
t
N
P
Examples
Consider the following activity measurement report (generated with the commands
shown):
display trace depth 512 <RETURN>
trace counting time <RETURN>
performance_measurement_initialize addr_ranges <RETURN>
performance_measurement_run 20 <RETURN>
performance_measurement_end <RETURN>
!perf32 | more
240
Mean of the standard deviations.
m
Table entry in Student's "T" table for a given confidence
level.
Number of traces in the measurement.
Mean of the means (i.e., mean sample).
m