3. You can run the same job multiple times to see what the
variability is for the particular job.
Agreed, that seems to be the only sensible solution, though not an
entirely satisfactory one.
Why would this be unsatisfactory? On a z9 processor, 1.7 billion machine
cycles (operations) can occur every second, for every engine. What makes
you think that you can even remotely begin to describe what "identical"
means for any given unit of work. The very nature of the interactions
between programs practically requires variability rather than
predictability. Even if you thought you could isolate each event, there are
many that will be unknown (not disclosed in the architecture), subject to
variation due to the last function performed, and unmeasurable. Why would
you think that absolutely single measurement repeatability was possible?
As I stated before, there will be variations, but it isn't random.
Similarly in all my years, it is no coincidence that most operations
departments were quite comfortable in knowing roughly how long any given job
was expected to run (I know there are/were exceptions). The reason they
knew this is because it was repeatable. The fact is that an "identical"
unit of work will operate around a range of values and can be measured.
Once measured, then the variation can be analyzed and a reasonable
probability can be defined to account for the work unit's run time.
If you want to know how a particular unit of work will be behave that you
need to come up with a probability distribution function and run it
repeatedly under controlled circumstances. This isn't simply some arbitrary
point, it is absolutely required to be rigorous. This concern about
applications is ill-conceived if it is thought that a one-time measurement
should produce reliable results. If an improvement occurs, then the
probability function will change and the results can be quantified.
In truth, I suspect most of this work hasn't actually been done anywhere to
either:
1. Define "identical"
2. Set up conditions to replicate as close to "identical" as possible the
circumstances for a unit of work
3. Measure the variation under controlled repeatable circumstances.
4. Measure the variation under varying loads. (which negates "identical").
Anyway .... my two cents
Adam
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
