Data Sampling Considerations

HPjmeter uses sampling, rather than exact counts, to minimize the overhead on applications. The accuracy of data collected by sampling increases with the number of samples. The time required to collect enough samples to get good accuracy varies, depending on the sampling frequency.

HPjmeter uses a confidence interval to give you a real-time view of the current accuracy of certain metrics.

For example, the Java Method HotSpots window shows the confidence interval numerically as a plus-or-minus percentage that appears just to the right of the measured value. A gray diamond underlies the text to show a graphical representation of the confidence level.

The horizontal positioning of each gray diamond represents the measured value in relation to the other measured values in the display. The width of the diamond represents the percent confidence interval. The real value measured lies somewhere within the range of confidence represented by the diamond.

During a live session, the width of the confidence interval shrinks over time as the statistical confidence in the value increases.

This example shows a high confidence in the results, denoted by the narrow gray diamonds in the display.

This example shows low confidence in the results, denoted by the wide gray diamond shapes.

Until the algorithm can show the data with confidence, the graphical confidence interval is displayed as a gray bar showing only relative relationship.