Previous Topic: 9.6.2.1 Control ParametersNext Topic: 9.6.2.3 Residual Analysis Report


9.6.2.2 Model Analysis Report


The Model Analysis Report in Figure 9-13 shows two models for
evaluation.  The best one-variable model includes only
MPPCPUTM as the independent variable, while there is only one
two-variable model to evaluate.

CA MICS Capacity Planner ANALYSIS OF MULTIVARIATE REGRESSION MODELS MODEL OF: CTLCPUTM BASED ON: MPPCPUTM BMPCPUTM --------INDEPENDENT--ELEMENTS-------- R**2 F P INTERCEPT NAME COEFFICIENT F P -------- ------- ----- --------- -------- ----------- ------- ----- 0.98 5806.11 .0001 3:06:54.3 MPPCPUTM 0.293982 5806.11 .0001 0.98 3286.01 .0001 1:14:09.0 MPPCPUTM 0.277912 2763.89 .0001 BMPCPUTM 0.170892 17.50 .0001


 Figure 9-13.  IMS Control Region Case Study Model Analysis Report

As we noted previously, on the first execution of the model
the analyst used the default Minimum R-squared improvement
value of 0.05, which resulted in a choice of the first model
(the one-variable model).  The re-execution of the analysis
with a Minimum R-squared improvement value of 0.001 resulted
in the two-variable model.  Both models show excellent values
of r-squared, F, and p, with the r-squared values for both
models less than 1% apart.

In a situation like this, it is hard to defend the choice of
one model over the other on the basis of statistical evidence
alone.  The analyst decided to use the two-variable model
since it takes into account both types of dependent regions
and therefore (in the analyst's opinion) passes the common
sense test with a marginally better score than does the
one-variable model.

The two-variable model indicates that the value of m1 is
0.277912 and the value of m2 is 0.170892.  This means that
each additional CPU second of MPP processor utilization will
generate an additional 0.277912 CPU seconds of control region
overhead.  Similarly, each additional CPU second of BMP
processor workload will generate an additional 0.17 CPU
seconds of control region overhead.