Performance Analyzer agent looks at the tivoli Data Warehouse where the historical data is stored, and does some calculations on input data.
The Input is the historical data collections from various agents. ( Like OS, DB2, VMware, ITCAM , SystemP, and Oracle are the only ones that are currently supported domains in 6.3.0.x )
The Output of this calculations is stored back into the Tivoli Data Ware house.
The PA agent can do analytical calculations using either the 1) Arithmetic , or the 2) Linear Trending methods.
1. Arithmetic - means - it can use user defined expressions and can produce a new meaningful result which may be useful to the end-user. ( a simple example maybe Max Memory - MinMemory of a server , OR Swapspace available - swap space used.
To do this : click on the PA icon and select OS -> Available memory Differential task ( which is an out-of-the box task ) and go to Input tab -> at the top you can enter your expressions.
2. Linear Trending : this uses the "Least Squares Regression" method. This method approximates a linear pattern of use, over time, for selected attributes based on the historical data ( or past data )
Now, LSR - is most commonly used mechanism to draw a line based on these values that will give an estimate of how the resource will perform over time in future.
e.g: How the CPU utilization may impact the server - if the current usage is continued for the next specified intervals - like next 7 days, 30 days and 90 days.