9. MULTIVARIATE REGRESSION FORECASTING › 9.6 Case Studies › 9.6.2 IMS Control Region Case Study

9.6.2 IMS Control Region Case Study

The IMS control region case study examines the IMS subsystem,
but the application is very similar in aim and execution to
the capture ratio case study.

OVERVIEW OF IMS

Think of an IMS system as a collection of address spaces (at
least two) that function together to process certain types of
batch and/or transaction-oriented workloads.  The address
spaces fall mainly into two categories: a control region and
dependent regions.  There is only one control region for an
IMS system (although with IMS Version 2, this distinction
becomes somewhat murky, since the DL/I and DBRC functions
previously contained within the control region have been
moved outboard to separate "dependent" address spaces.  We
can ignore this exception for the moment and assume that all
IMS control functions are contained in one place).

The control region provides the control functions necessary
to keep the IMS system running as a unified database/data
communications system: integrity, recovery, data sharing,
dispatching, queuing, etc.  The dependent regions are the
address spaces that process the user-generated workload, both
batch job and transaction.  Dependent regions execute under
the control and guidance of the IMS control region in much
the same way that all executing tasks within a processor run
under the control and guidance of the operating system.

For our case study, we are looking at two types of dependent
regions: the BMP (Batch Message Processing region), and the
MPP (or Message Processing Program).  The BMP can be
described as a batch job that runs under the guidance and
control of the IMS control region, while an MPP is a task
(job or started task) that processes transactions normally
originating at terminals and returns output to terminals.

Just as all executing tasks within a processor incur overhead
due to the control functions exercised on their behalf by the
operating system, so do all IMS BMP and IMS transactions
incur control region overhead.  The more transactions
processed, or BMPs run, the more CPU time that is consumed by
the control region on their behalf.  In large IMS systems,
the amount of CPU time used by the control region alone can
be quite significant (30 to 50% of a processor), and this
utilization must be taken into account in the capacity
planning process.

If, for example, the capacity planner of the XYZ Corporation
knows that the IMS transaction load will increase by a
certain amount, and the IMS BMP workload will increase by a
certain amount, then the capacity planner still must
determine how much increase will occur in control region
utilization before adequately predicting the growth
requirements for the IMS system.

To do this, the capacity planner guesses that the problem of
determining the control region utilization from the MPP and
BMP region utilizations could probably be estimated by a
multilinear regression model.  Assume that the control region
has been established in its own performance group in the
IEAIPS01 member of SYS1.PARMLIB.  Furthermore, assume that
the MPP regions have been established in one (or more)
separate performance groups, and that the BMPs also execute
in their own performance group(s).  The key point is that the
control, MPP, and BMP regions' utilization may all be
isolated from one another by performance group assignments.

The capacity planner can then establish a single capacity
planning database resource element file to archive
utilization statistics from the CA MICS SCPPGA File.  This
resource element file is called IMS.  The CPU time variables
in this file are called CTLCPUTM, MPPCPUTM, and BMPCPUTM
(denoting aggregated CPU time for all control regions,
message processing regions, and batch message regions,
respectively).

The capacity planner reasons that if MPP processor use
increases, there is a strong likelihood that control region
use will increase by a proportional amount.  The same is also
true of BMP use.  Therefore, a model of control region use
would be:


    CTLCPUTM = m1 * MPPCPUTM + m2 * BMPCPUTM + b      (Eqn 9)

You could interpret this formula by saying that each
additional CPU second of MPP processor use incurs m1 CPU
seconds of control region use, while each additional second
of BMP use incurs an additional m2 seconds of control region
use.

The interpretation of the value b is similar to that of the
capture ratio case study (Section 9.6.1).  This is a "noise"
factor, or the amount of processor busy time used by the
control region during a no-load period.  In other words, the
control region uses this amount of CPU time just searching
for work to do.  However, as in the case of the capture ratio
example, you must take care in evaluating the value of b.
See Section 9.6.1 for a discussion on the caution needed when
predicting outside the range of historical data.

The next sections explore the results of this model:

    1 - Control Parameters
    2 - Model Analysis Report
    3 - Residual Analysis Report