14. WORKLOAD FORECASTING › 14.6 Case Studies › 14.6.1 Processor Usage Study

14.6.1 Processor Usage Study

This case study presents an analysis of processor busy time
using a set of primary workloads to drive the study.  It
assumes that a preliminary analysis using the Resource
Component Analysis application has been performed using the
same Forecast and Analysis files as are input to this study,
and that a subset of workloads that best represent the
processor usage has been identified.

In performance analysis and capacity planning, the basis of
any study is the examination of measurement data.  In this
discussion, assume that the system being measured is a
processor running the MVS operating system (although the
underlying theory is independent of the type of operating
system).  In an MVS system, basic measurement data is usually
derived from SMF and RMF data, which you store in the CA MICS
Batch and Operation Analyzer (SMF) and Hardware and SCP
Analyzer (RMF) areas respectively.  For this case study we
will narrow our discussion to RMF data and consider the
relationships between data stored in the WLM Service Class
Resource Consumption (WLMSEC) File and the CPU Processor
Activity (HARCPU) File.


DEVELOPMENT OF CONCEPTUAL MODEL

Although a typical system executes many transactions across a
number of workloads, generally speaking, there will be a
subset of workloads. For example, some small number of
workloads that tend to drive the processor busy.  Some
systems may even be devoted primarily to a single workload
such as a dedicated IMS or CICS system, but that is not the
normal situation.  Although you could perform capacity
planning using all workloads, such an approach requires more
effort on the part of the analyst and would likely not yield
more accurate results.  Generally, it is better to plan on
the primary workloads and ignore smaller components.

In our case study, four workloads have been determined to be
the most "important", statistically speaking:  BATCH, ONLINE,
DATABASE and TSO. There were actually eight workloads
represented in the input Analysis file, but four were
discarded for the purposes of this case study. One
interesting point was noted while preparing this case study.
The TSO workload was ultimately included even though it was a
minor user of processor resources, considerably less than
some of the workloads that were discarded.  The preliminary
analysis using the Resource Component Analysis application
assigned a higher Importance value to this workload than the
other minor players.  While the exact reason for this
observation is unknown, Neugents technology employed by this
application uses a nonlinear process to analyze data and
therefore can discover relationships among data that are not
as apparent when using other statistical methods of analysis.
It is quite possible that the small TSO workload is employed
mainly to "drive" one of the other larger workloads, such as
the DATABASE workload.  This demonstrates that there may be a
indirect relationship between the TSO and DATABASE workloads,
despite the rather low resource usage.

APPLICATION OF CONCEPTUAL MODEL

The model was developed and the training phase executed. The
Training Phase Summary statistics showed that out of 53 weeks
of input history data, the application selected 40 cases for
training, yielding a model "goodness of fit" of 0.9960,
meaning that the model prediction could account for over 99
percent of the forecasted values. Additionally, the average
model error was less than 1 percent, indicating that the
model was performing well.

A period of 13 weeks was chosen for the Forecast Length and
the model output again demonstrated some interesting
characteristics.  In particular, the model predicted a nearly
"flat" growth rate, not especially surprising given the
nature of the input data. But the predicted values actually
first curve upwards slightly and then curve downwards towards
the end of the forecast. This clearly shows the nonlinear
nature of the processor data and ability of Neugents
technology to establish relationships among such data.

The rest of this section discusses the following topics
relating to this case study:

     1 - Control Parameters
     2 - Input Data Descriptive Statistics
     3 - Training Phase Summary
     4 - Forecasting Phase Analysis Report
     5 - Forecasting Phase Analysis Graph