When the parameters presented on the panels in Section 3.1.5.2 are executed, the reports shown in Figures 3-7, 3-8, 3-9, and 3-10 are produced. Since the purpose of this run is to validate the model, the Job Input Queue Time Report and Job Class Throughput Report are the most important because they give a comparison of the actual and simulated data. The other two reports are produced from the simulation results. Each report is discussed next. INITIATOR BUSY REPORT The principal use of the Initiator Busy Report shown in Figure 3-7 is to see how busy each initiator is during the simulation period. In addition, the report provides a breakdown of the workload by class for initiators that serve several classes. Initiators which are lightly used (under 25 percent) are candidates for elimination. Figure 3-7 shows that for the initiator structure used by the sample system, none of the initiators are heavily loaded; indeed, many are very lightly loaded. This same initiator structure is used in the application strategy, where changes in the structure are tested. The fact that so many initiators are lightly loaded is central to the changes proposed. When studying this report, also review the breakdown of occupancy time by class for heavily loaded initiators. This breakdown can be useful in detecting situations where one class is being blocked by other classes and, therefore, is not being adequately served by the initiator structure. This in not a consideration for the sample system, because each initiator serves a single class.
INITIATOR BUSY REPORT
SYSTEM = TEST RUN DATE = JAN 8, 1988
MODEL PARAMETERS
START DATE/TIME END DATE/TIME
07JAN88:07:00 07JAN88:17:00
PRIORITY AGING
PRILOW PRIHIGH PRIRATE
5 11 24
INIT 1 CLASS: D
CLASS JOBS STARTED % BUSY
------- ------------ ------
D 23 77.7
INIT 2 CLASS: D
CLASS JOBS STARTED % BUSY
------- ------------ ------
D 22 64.7
INIT 3 CLASS: D
CLASS JOBS STARTED % BUSY
------- ------------ ------
D 26 48.4
INIT 4 CLASS: D
CLASS JOBS STARTED % BUSY
------- ------------ ------
D 14 30.8
INIT 5 CLASS: D
CLASS JOBS STARTED % BUSY
------- ------------ ------
D 17 19.8
INIT 6 CLASS: D
CLASS JOBS STARTED % BUSY
------- ------------ ------
D 4 32.9
INIT 7 CLASS: D
CLASS JOBS STARTED % BUSY
------- ------------ ------
D 6 21.0
...etc...
Figure 3-7. Case Study - Initiator Busy Report
JOB QUEUE SIZE REPORT
The Job Queue Size Report shown in Figure 3-8 serves as a
check of the information in the Job Input Queue Time Report
by showing fluctuations in the job queues over the simulation
period.
The report for the sample system shows the following:
o The start of the simulation was a period of little or
no batch activity (no jobs arrived in the system).
This is desirable; it means that the simulation will
be more accurate.
o Classes A, B, C, and D were active. Classes F and W
had little activity (these are only used for special
purposes on the system being studied).
o No class was subject to particularly large
fluctuations in workload arrival rates.
o Several classes show spikes in arrivals (e.g., 15:00
for class B), but the spikes are generally worked off
in the next few periods. Class D shows some
sluggishness in this regard and should be investigated
further.
The Job Input Queue Time and Job Class Throughput Reports are
of particular interest because they contain both actual and
simulated data. In addition, both reports contain data on
queue times (one in a plot and the other as an average). The
two reports are discussed separately below, followed by some
comments on comparing the information in the two reports.
JOB QUEUE SIZE REPORT
SYSTEM = TEST RUN DATE = JAN 8, 1988
MODEL PARAMETERS
START DATE/TIME END DATE/TIME
07JAN88:07:00 07JAN88:17:00
PRIORITY AGING
PRILOW PRIHIGH PRIRATE
5 11 24
-------------------------------------------------------------------------------
| | CLASS A | CLASS B | CLASS C | CLASS D | CLASS F | CLASS W |
| TIME|MAX END NO|MAX END NO|MAX END NO|MAX END NO|MAX END NO|MAX END NO|
| |QUE QUE ARR|QUE QUE ARR|QUE QUE ARR|QUE QUE ARR|QUE QUE ARR|QUE QUE ARR|
|-----+-----------+-----------+-----------+-----------+-----------+-----------|
| 7:00| 2 1 2| 1 0 1| 3 0 6| 2 1 2| . . .| 1 0 4|
| 7:30| 1 1 0| 1 0 2| 2 2 2| 2 2 2| . . 0| 1 0 1|
| 8:00| 2 1 3| 1 0 7| 3 3 1| 4 2 11| . . 0| 0 0 0|
| 8:30| 2 1 5| 8 0 15| 10 4 21| 4 2 11| . . 0| 0 0 0|
| 9:00| 4 3 9| 6 5 13| 8 7 7| 4 2 2| 1 0 1| 0 0 0|
| 9:30| 4 3 3| 18 18 20| 8 8 2| 12 11 15| 1 0 1| 0 0 0|
|10:00| 11 10 19| 27 26 17| 17 16 10| 16 15 13| 0 0 0| 0 0 0|
|10:30| 12 11 11| 38 37 14| 21 20 10| 20 19 7| 0 0 0| 0 0 0|
|11:00| 16 15 24| 45 38 17| 25 20 8| 33 31 21| 0 0 0| 0 0 0|
|11:30| 20 18 16| 43 38 11| 24 23 6| 44 43 15| 0 0 0| 0 0 0|
|12:00| 21 5 39| 40 13 18| 36 19 30| 44 32 3| 1 0 1| 1 0 1|
|12:30| 7 5 10| 24 21 30| 22 19 7| 33 32 3| 1 0 1| 0 0 0|
|13:00| 7 5 26| 27 23 20| 21 19 10| 34 32 5| 0 0 0| 0 0 0|
|13:30| 9 8 16| 36 35 30| 23 23 7| 34 33 5| 0 0 0| 0 0 0|
|14:00| 17 14 19| 41 38 17| 25 24 8| 36 34 10| 0 0 0| 0 0 0|
|14:30| 25 24 28| 51 50 28| 36 36 15| 37 35 9| 0 0 0| 0 0 0|
|15:00| 42 40 35| 92 89 55| 48 48 20| 39 38 12| 0 0 0| 0 0 0|
|15:30| 42 35 21| 91 86 15| 55 54 12| 54 54 18| 1 0 2| 0 0 0|
|16:00| 36 25 20| 92 91 9| 63 63 12| 67 65 17| 0 0 0| 0 0 0|
|16:30| 26 20 8| 93 82 8| 63 61 7| 69 67 10| 0 0 0| 0 0 0|
|17:00| 21 17 1| 82 82 0| 61 61 0| 67 67 0| 0 0 0| 0 0 0|
-------------------------------------------------------------------------------
Figure 3-8. Case Study - Job Queue Size Report
JOB INPUT QUEUE TIME REPORT
The Job Input Queue Time Report shown in Figure 3-9 is
organized by queue time. Strong agreement of the simulated
and actual distributions in the report indicates that the
model accurately simulates the system being studied.
Discrepancies indicate problems in the model for particular
classes or for the entire system.
We can make the following comments about the plots in the
example:
o The plot shows a high tail for the actual queue time.
This usually reflects some external interference such
as an operator hold. This lengthens the actual queue
time but is not reflected by the simulation. If the
tails of the distributions diverge sooner than about
the 95th percentile, you should investigate the causes
and be cautious in using the simulation results.
o The number of observations hidden are caused by two
factors: points at which the two plots coincide and
points at which several observations occur for one of
the plots (either actual or simulated). The large
number of hidden observations means that the
simulation closely reflects the actual system.
JOB INPUT QUEUE TIME REPORT 13:05 FRIDAY, JANUARY 8, 1988 ACTUAL QUEUE TIME = A SIMULATED QUEUE TIME = _ JOB CLASS=D PLOT OF JOB_AQTM*ACT_DONE SYMBOL USED IS A PLOT OF JOB_SQTM*SIM_DONE SYMBOL USED IS _ | 6:56:40 + | A | A | | | 5:33:20 + I | N | P | U | T | A 4:10:00 + Q | U | E | U | E | 2:46:40 + T | _ I | _ M | A_ E | __ | __ 1:23:20 + __ | ____ | ___ A AAAAAA | _ AAA | ___ AA | _______________________________________________________________________________________ AAAA 0:00:00 + AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA -----+----------+----------+----------+----------+----------+----------+----------+----------+---------+----------+----- 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0 PERCENTAGE OF JOBS SELECTED NOTE: 119 OBS HIDDEN
Figure 3-9. Case Study - Job Input Queue Time Report
JOB CLASS THROUGHPUT REPORT
As with the Job Input Queue Time Report, strong agreement of
the simulated and actual data in the Job Class Throughput
Report, shown in Figure 3-10, is desirable, because it means
that the simulation closely reflects the actual system.
We can make the following comments about the report:
o Classes with jobs that are very short may not be well
simulated by the model, depending on the value
specified for the simulation clock step.
In our sample, the simulation clock step value is 30
seconds. As the report shows, the actual times for
job classes F and W are not close to the 30-second
value; they are not well simulated by the model.
o The average queue time, while reasonable for some
classes, is not in particularly close agreement
overall for the simulated and actual data. This
indicates that further investigation of the system
under study is necessary.
An analysis of other variables in the BATJOB File
showed JES priority aging taking place at a rate which
did not agree with the definition used in the model.
In addition, the average is a poor descriptor of the
distribution of queue times for this data. This is
discussed below.
COMPARISON OF QUEUE TIME AND THROUGHPUT REPORTS
Since both the Job Input Queue Time and Job Class Throughput
Reports contain information describing the input queue time,
it is important to compare the data in the two reports.
In our example, the comparison is striking. The throughput
report shows queue times in the range of 0 to 30 minutes.
The plots in the queue time report reveal that few, if any,
jobs have queue times near the average. The majority of jobs
have little or no queue time at all. Those that do have a
queue time have times much longer than average. In short,
the distribution is quite skewed; the average is not a good
descriptor of the jobs being studied.
In general, if the value for the coefficient of variation is
greater than 3 on the Job Class Throughput Report, the
average input queue time may be an unreliable measure due to
the high degree of variability of the data. Further
investigation of the system under study may be needed prior
to drawing conclusions.
JOB CLASS THROUGHPUT REPORT
SYSTEM = TEST RUN DATE = JAN 8, 1988
MODEL PARAMETERS
START DATE/TIME END DATE/TIME
07JAN88:07:00 07JAN88:17:00
PRIORITY AGING
PRILOW PRIHIGH PRIRATE
5 11 24
------ ACTUAL ----- ---- SIMULATED ----
JOBS AVG WAIT CV JOBS AVG WAIT CV
A 313 0:11:05 3.1 297 0:13:59 2.9
B 306 0:29:08 2.8 265 0:25:09 2.4
C 147 0:11:01 3.3 140 0:12:54 3.5
D 123 0:16:48 3.7 120 0:11:25 2.5
F 6 0:00:07 0.2 6 0:00:00 0.0
W 1 0:00:10 0.0 1 0:00:00 0.0
Figure 3-10. Case Study - Job Class Throughput Report
|
Copyright © 2014 CA.
All rights reserved.
|
|