Figure 9-10 shows the Model Analysis Report for this case
study. The model that was chosen by the stepwise regression
process is the last one listed. It has an r-squared value of
.97 and acceptable F and p values, so that the first test is
passed.
The only variable that gives some concern is the BATCPUTM
independent variable, which has a somewhat high value for the
p statistic. You can interpret the p statistic for each of
the independent variables as the probability that the
coefficient (in this case 1.11713) is 0. While a 3.45%
probability is not really very large, it is considerably
higher than its neighboring independent variables. In this
case, the cause of this somewhat higher p statistic can be
explained by the fact that the batch workload is a much
smaller workload than the others, and therefore variances in
the workload may be harder to account for in a linear model.
In any event, you should not be very concerned with this p
statistic, since the capture ratio for batch that results in
this model seems very reasonable.
Recall from the discussion in Section 9.6.1 that the capture
ratio is equal to the inverse of the regression coefficient,
m. The capture ratio values resulting from this model are:
TSO : 1 / 1.42295 = 0.70
BATCH : 1 / 1.11713 = 0.90
IMS : 1 / 1.32701 = 0.75
CICS : 1 / 1.28791 = 0.77
Finally, note that the b value for this model is negative,
violating the common sense test. So why do we consider this
final model an acceptable one?
Recall the discussion of the introduction in Section 9.6.1
concerning low utilization effects. Examination of the input
historical data series (see Figure 9-11) reveals that all of
the historical data points occur at moderately high to very
high CPU utilization rates. Consequently, the predicted
value for b is meaningless, since there are no historical
data points to guide the regression when the load is light.
Therefore, you can determine that the value of b in this
case, even though it is negative, is unimportant and
uninteresting.
CA MICS Capacity Planner ANALYSIS OF MULTIVARIATE REGRESSION MODELS MODEL OF: TOTCPUTM BASED ON: TSOCPUTM BATCPUTM IMSCPUTM CICCPUTM --------INDEPENDENT--ELEMENTS-------- R**2 F P INTERCEPT NAME COEFFICIENT F P -------- ------- ----- --------- -------- ----------- ------- ----- 0.94 1684.11 .0001 44:16:57.0 CICCPUTM 2.96864 1684.11 .0001 0.95 977.72 .0001 29:12:42.3 IMSCPUTM 0.594775 16.58 .0001 CICCPUTM 2.72044 895.89 .0001 0.96 1286.48 .0001 -2:19:38.6 TSOCPUTM 2.88379 1195.40 .0001 IMSCPUTM 1.63519 222.39 .0001 0.97 1127.97 .0001 10:57:58.3 TSOCPUTM 1.77499 71.77 .0001 IMSCPUTM 1.14392 77.90 .0001 CICCPUTM 1.13065 31.88 .0001 0.97 877.25 .0001 -3:58:16.7 TSOCPUTM 1.42295 29.20 .0001 BATCPUTM 1.11713 4.60 .0345 IMSCPUTM 1.32701 74.89 .0001 CICCPUTM 1.28791 37.62 .0001
Figure 9-10. Capture Ratio Case Study Model Analysis Report
|
Copyright © 2014 CA.
All rights reserved.
|
|