11. WORKLOAD CHARACTERIZATION › 11.6 Case Studies › 11.6.1 Batch Job Class › 11.6.1.3 Report Interpretation › 11.6.1.3.5 Analysis of the ALL_OBS SAS Data Set

11.6.1.3.5 Analysis of the ALL_OBS SAS Data Set

The final step of the job class identification process
involves some decisions based on the cluster descriptions and
the analyst's experience.  Although the clusters developed in
the example provide an excellent representation of the
system's workload, there are too many clusters to define a
job class structure with a class for each structure.
Providing a minimum number of alternative job classes would
simplify the job class selection process.  Therefore, after
analyzing the data from the cluster centroids and the
population statistics reports (see Figures 11-26 and 11-28),
we proposed four job classes.  These classes are shown in the
following table.


                                               CLUSTERS
    CLASS   JOBMXNTA   JOBTCBTM    JOBNLR     REPRESENTED
    =====   ========   ========   ========   =============
      1        0           5        2000          7
      2        0          60        5000       2,4,9,6
      3     >=1, <=2      60       20000     1,3,5,8,10,11
      4      unlim       unlim     unlim       outliers


A SAS program was written to evaluate the proposed classes.
The program attempts to assign each job to the classes
(tested in ascending order of size) until a match is found.
The results from the execution of this program are shown in
Figure 11-29.

As you can see in the figure, 10,722, 4,155, 1,489, and 2,267
jobs fell into the four classes.  These values correspond to
58, 22, 8, and 12%, respectively, of the system's workload.
To test the sensitivity of these job class limits, we should
repeat the analysis varying the job class limits by + or -5%.
If the results of these analyses do not significantly change
the percentage of the workload represented by each of the
classes (that is, more than 2 or 3% of the jobs change
classes), then you can be confident that none of the job
classes bisect a resource consumption pattern in the
workload.

                    1          DATA _NULL_;
                    2          SET JOBFILE.ALL_OBS END=EOF_SW;
                    3          RETAIN C1 C2 C3 C4 0;
                    4
                    5            IF JOBMXNTA<=0 & JOBTCBTM<=05 & JOBNLR<=02000 THEN DO;
                    6              C1+1;
                    7              GO TO T_EOF;
                    8            END;
                    9            IF JOBMXNTA<=0 & JOBTCBTM<=60 & JOBNLR<=05000 THEN DO;
                    10             C2+1;
                    11             GO TO T_EOF;
                    12           END;
                    13           IF JOBMXNTA>=1 & JOBMXTAP<=2 & JOBTCBTM<=60 & JOBNLR<=20000 THEN DO;
                    14             C3+1;
                    15             GO TO T_EOF;
                    16           END;
                    17           C4+1;
                    18         T_EOF:
                    19           IF EOF_SW THEN DO;
                    20             PUT C1= / C2=/ C3=/ C4=;
                    21             P1=C1/(C1+C2+C3+C4)*100;
                    22             PUT P1=;
                    23             P2=C2/(C1+C2+C3+C4)*100;
                    24             PUT P2=;
                    25             P3=C3/(C1+C2+C3+C4)*100;
                    26             PUT P3=;
                    27             P4=C4/(C1+C2+C3+C4)*100;
                    28             PUT P4=;
                    29           END;

                    C1=10722
                    C2=4155
                    C3=1489
                    C4=2267
                    P1=57.45
                    P2=22.30
                    P3=7.99
                    P4=12.17

                    NOTE: THE DATA STATEMENT USED 1.76 SECONDS AND 384K.
                    30         RUN;
                    NOTE: SAS USED 384K MEMORY.

                    NOTE: SAS INSTITUTE INC.
                          SAS CIRCLE
                          BOX 8000
                          CARY, N.C. 27511-8000

 Figure 11-29.  Sample ALL_OBS Analysis Program