Report Calculation Principles And Definitions; The Difference Between Average And Peak Average; The Manner In Which Trends Are Calculated - IBM PERFORMANCE MANAGEMENT FOR POWER SYSTEMS - GRAPH REFERENCE DOCUMENT Reference

Report Calculation Principles and Definitions

The Difference Between Average and Peak Average

By default, operating statistics are summarized and averaged every 15 minutes under normal
operation. Average and peak average statistics are collected and presented in the PM for Power
Systems reports as:

Average
The average figures represent the average of all the 15 minute utilization summaries,
across each shift of each day included in the reporting period.

Peak Average
The peak average figures represent the averages of the utilization summaries for the
highest two hour period for each shift of each day of the reporting period. Usually the
performance constraints occur during this time period.

The Manner in which Trends are Calculated

The PM for Power Systems trend calculations are made using
based on up to twelve months of historical data. Predictions are still produced if less than one
year's worth of data is available; however a minimum of three months data is required.
Note: The accuracy of these predictions depends on the information that is available.
The use of regression analysis in a rapidly changing environment can tend to disguise sudden
changes away from the trend. PM for Power Systems includes trend lines for processor and disk
utilization based on the previous three and six months of historical data for this reason, in addition
to twelve months. This allows you to more easily see rapid changes in utilization over a short
period of time and to evaluate your growth and business needs based on your most recent
information.
Normalization is the basis on which PM for Power Systems reports are calculated. It takes into
account configuration changes, either to the system or to a logical partition (LPAR) during the
reporting period.
The basis used for all historical and predicted data is the relative commercial processing
workload (CPW) in the case of IBM i, and rperfs in the case of AIX. All models of Power Systems
have a performance value that represents the relative amount of processing which can be
performed by that model. Using this normalized value to plot workload demand allows you to see
a more accurate reflection of your workload changes even if a processor upgrade has taken
place. It is also important in a logical partitioning environment to see the amount of processing
power being used across (dynamic) partitions.
Note: The percentages shown on the reports are based on the configuration of the system or
LPAR at the end of the last 15 minute period of the month of the reporting period. This can lead to
reports showing in excess of 100% utilization when the resource assigned at this time is less than
during other periods of the reporting period.
10 PM for Power Systems Graph Reference Document
linear regression analysis
and are