So, my question is, suppose we are in a high CPU use situation. In fact,
the CPU is running 100% and some jobs are not receiving any CPU.
Everything is running behind in service. That is, the PI for all batch
work is >1 . I want to ensure that non-production work will receive CPU
cycles. What is really wanted to to say that production batch will
receive 60% of the cycles, Model office about 20% and test about 10%.
These percentages are not of the entire CPU, but only of the CPU "left
over" after works such as production CICS, JES2, NET, etc get what they
need.
I'm confused by your statement regarding "everything is running behind in
service". It seems to me that the first question that needs to be answered
is "What are the real-world objectives the workload has to achieve?" A PI >
1, simply means that your defined goal hasn't been met, but it says nothing
about the real-world experience seen by the end-user. In other words, don't
specify response time goals of 0.5 seconds when 0.7 seconds would do.
Your SYSTEM and SYSSTC service classes will do fine by themselves. After
that it is important to ensure that the goals for the other service classes
reflect reasonable real-world objectives, rather than simply being numbers
that look good on performance monitors. If the higher importance level
service class goals aren't too aggressive, then (in the absence of capacity
problems), there will be enough resources left to ensure that everyone gets
a small piece.
Using Resource Groups is NOT a good idea unless you're prepared to have
higher importance work miss its goals in order to honor the minimums. An
approximation of the lower velocities can be calculated as:
{[(Total number of LPs) * (Percent avail)] / (Number of competing tasks) }
* 100
The first term will approximate how much CPU resource is available after the
higher importance work has taken its share, while the number of competing
tasks is an approximation of how many work units are expected to be
competing simultaneously.
In any case ... 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
