RE: Measuring Qinter performance & page faulting

["Andy Nolen-Parkhouse" <aparkhouse@xxxxxxxxx>]

Gerald,

So much in your question depends upon what problem you are trying to
solve.  The best and most accurate measurement of interactive performance
will always be happy users.  They will have a mental baseline of what is
acceptable and will let you know when that baseline is exceeded.  As you
examine the system over time and with various loads, you should develop a
feel for acceptable levels of faulting/paging.  The formula you cited
should give you an indication of when you are getting close to a
performance-throttling threshold, your users will let you know when you
reach it.

If you work in a very empirical shop, then such formulas are valuable for
demonstrating to management that you need to upgrade or tune the machine.
You can cite IBM-supplied formulas and measurements from your own machine
over time to justify your opinions.  If you work in a less formal shop,
you will probably know when memory becomes an issue without all of the
technicality.

Are you trying to solve a particular problem?  Are you using shared
pools?  If you're using shared pools, have you isolated the interactive
users to a single pool without other types of jobs?  Is performance
adjustment turned on?

If you are allowing performance adjustment to tune your system, then the
amount of work in other pools can have a significant effect on your
interactive performance.

The Work Management manual has not really been produced as a full edition
since V4R5.  While the basic information is probably accurate, the IBM
InfoCenter may have more current suggestions.

Regards,
Andy Nolen-Parkhouse

> Are there any other methods that would be a better indicator of qinter
> subsystem performance relative to page faulting? Is the work management
> manual still a reliable guideline?
> 
> Thanks
> 
> Gerald Kern