Determine the appropriate CAB parameters to improve
the CAB performance.
About this task
The CAB performance can be considered satisfactory when
the following conditions are met:
- The number of OSAM blocks, OSAM LDS CIs, or ESDS CIs read in chained sequential
mode is much greater than the number of blocks or CIs read in direct mode.
- The percentage of unreferred-to sequential buffers is below 15
or 20%.
- Elapsed time is substantially smaller with CAB than with BB.
- The amount of buffer space that is used is acceptable.
If you did not get satisfactory results by running your IMS High Performance Unload job with
the default CAB parameters, you can attempt tuning.
Procedure
The following steps describe a possible approach to
tuning CAB's reference pattern analysis logic:
- Run your application program or the FABHTEST utility on
a dedicated system with CAB and with the default values for the CAB
parameters.
Activate the machine-readable buffer handler
trace for future FABHBSIM runs. (This increases the processor time
slightly; bear this fact in mind when you compare processor times.)
Note: Performance
tests with databases that contain fewer than 300,000 segments are
not reasonable because CAB buffer handler has a high initialization
overhead.
If you code 'CABSTAT YES' in the HSSROPT data set,
extensive buffer handler statistics are printed at the end of the
job step on the HSSRSTAT data set. See CAB Statistics report and Data Set I/O Statistics report.
Tip: After you have tuned the buffer handler, you can remove
the CABSTAT control statement to reduce the content of the report.
If no CABSTAT control statement is coded, or 'CABSTAT NO' is specified
in the HSSROPT data set, only the summary page of the CAB Statistics
is printed for each buffer pool.
- Run the same program with BB on a dedicated system by coding
the BUF control statement in the HSSROPT DD.
- Check the results of both runs. First compare the elapsed
time.
Note the number of sequential buffers not referred
to, and the number of blocks read in chained sequential mode and direct
mode.
If your database is well organized, CAB should show
excellent elapsed time figures. Unless you want to reduce the buffer
space, no further tuning is needed.
If your database is poorly
organized, CAB might show moderate or disappointing results.
If
the improvements are moderate—that is, the elapsed time of the
CAB run is between 65% and 80% of the BB run—you should investigate
whether the database reorganization interval is appropriate and whether
the free space specifications are appropriate.
You can also
investigate the increase of the NBRSRAN parameter and the tuning of
the REFT4 parameter to decrease a little more the elapsed time.
If
the results are disappointing—that is, if the elapsed time of
the CAB run is not below 80% of the first run—the best thing,
probably, is to reorganize the database more frequently to improve
the specification of free space.
- If you decide to rerun with modified CAB parameters, you
can use the FABHBSIM utility to compare the HSSRSTAT statistics of
the original run with those of the new run.
The FABHBSIM
utility enables you to observe the effect that changes to parameters
on HSSRCABP control statements have on buffer handler performance.
FABHBSIM can be used for simulations of both CAB and BB.
The
statistics to be compared are the number of direct I/Os, the number
of chained sequential I/Os, the number of sequential buffers note
referred to, and the timing statistics.
Restriction: FABHBSIM
cannot be used for HALDBs.