11. WORKLOAD CHARACTERIZATION › 11.6 Case Studies › 11.6.1 Batch Job Class › 11.6.1.3 Report Interpretation › 11.6.1.3.2 Cluster Centroids Report

11.6.1.3.2 Cluster Centroids Report

The Cluster Centroids Report provides a summary of the
patterns that are identified by the clustering procedure.  A
sample Cluster Centroids Report is shown in Figure 11-26.
This report allows you to determine the characteristics of
the patterns that are identified.

                                                 CLUSTER CENTROIDS

                           CLUSTER  JOBMXNTA  JOBTCBTM  JOBNLR    RADIUS  NO ASSIGN  PCT
                           -------  --------  --------  --------  ------  ---------  ---
                              1          0.0   0:00:05    3450.7    2.00      145      7
                              2          1.0   0:00:03     277.8    1.00      163      8
                              3          0.0   0:00:57      90.9    3.00       68      3
                              4          1.0   0:00:23      64.4    2.50       56      3
                              5          0.0   0:00:14     181.8    1.00      156      8
                              6          2.0   0:00:03     209.3    3.00       58      3
                              7          0.0   0:00:01     267.7    1.00     1172     59
                              8          0.0   0:00:07    8165.7    3.00       54      3
                              9          1.0   0:00:03    3950.7    2.50       20      1
                             10          0.0   0:00:11   12211.6    3.00       81      4
                             11          0.0   0:00:36     261.4    2.00       27      1

  Figure 11-26.  Cluster Centroids Report

On the sample, consider the values shown for CLUSTER in the
report.  Cluster 7 represents approximately 59% of all jobs
that executed during the seven-day period.  The cluster
centroid represents a job that uses no tape drives, uses one
second of CPU time, and prints approximately 270 lines.  The
radius of the cluster is 1.0 standard deviations.  Using the
trimmed standard deviation values from the example in Figure
11-15, we can determine the bounds of the jobs assigned to
the cluster.  Consider the following table for cluster 7:


                                        LOWER    UPPER
     FEATURE     CENTROID    STD DEV    BOUND    BOUND
     ========    ========    =======    =====    =====
     JOBTCBTM        0:01       0:11    -0:10    +0:12
     JOBMXNTA           0       0.39    -0.39    +0.39
     JOBNLR           267       2339    -2072    +2606


By multiplying the positive and negative values of the
cluster's radius times by the standard deviation for each of
the features, you can determine the possible bounds of the
cluster.  Note that you may allocate tapes only in integer
values and that jobs cannot consume negative resource.  Also,
note that the minimum value of CPU time observed in the
distribution was .01 second.  This simplification yields the
following table:


                                        LOWER    UPPER
     FEATURE     CENTROID    STD DEV    BOUND    BOUND
     ========    ========    =======    =====    =====
     JOBTCBTM        0:01       0:11     0:01    +0:12
     JOBMXNTA           0       0.39        0        0
     JOBNLR           267       2339        0    +2606


Although the second table still indicates that the resulting
cluster has some positive variability, the closeness of the
cluster's mean values to the minimum values (from the
Descriptive Statistics Report, Figure 11-27) for the
distribution indicates that the number of extreme values is
small.