How the confidence in Performance Analyzer is calculated.

This blog explains how the confidence in PA is being calculated.

Often times you will see the value varying even though you have good range of data being calculated.

The confidence value depends on the values in the TDW for the analytical task being calculated, and if the values are varying then the confidence will proportionately vary.
Higher the variance - lower the confidence.
Likewise - if the values are close to each other - then the confidence will come up close or higher than 90%/

( the other way I look at it is, if I know that the values are same or approximately the same each time the reading was taken and If I can select that range in the PA config window - then I can get a good confidence value , but for this I need to know what are the values in the Tivoli Data Warehouse )

Below is my experiment on Average Memory Utilization Analytical task for 11 days and is captured below on how I was able to increase the confidence by hand editing or simulating good data.
This is not dependent on the number of samples - but on the values.

--

Before:

I have 11 day worth of data for the Average Memory free, with varying values - and the calculated confidence was 25, since the "average memory free" was varying as shown below.



The snap shot below is the TDW data contents, The column on the left is the day representation split as follows 1 - 13 - 09 - 06 -23-date,  
( i.e Sep 6th 2013 and upto Sep 18th 2013. )  




--

After:

Now, let's change the value to a constant for all the days - by hand -editing the value in the TDW to 369.45 and watch that the Confidence goes to 100.



Watch the confidence jump to 100



Some supporting notes :
This was tested on RHEL 5. and DB2 housing the TDW, and all PA analytical tasks were running
To enable a quicker visual - I had to turn the analytical task to run every minute and watched the <ITM LOGS>kpacma.log  so that it would refresh faster.