RE: Measuring Qinter performance & page faulting

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

Gerald,

In your previous posting, you said that the formula from the manual
indicated that if per-transaction faulting/paging exceeded 20, then more
storage might be required.  Where is your co-worker getting his number
20?  If it is from the WRKSYSSTS display and doesn't account for
transaction volume, then I would agree with you that the measurement
needs a little more rigor.

As systems become more and more powerful, some of the rules of thumb need
to be updated.  If you were to look at an older Work Management manual,
you would find that IBM listed different acceptable faulting rates for
different models; the more powerful the machine, the higher the
acceptable level of faulting.

IBM is assuming that faulting will occur and that the goal of tuning is
to ensure that a transaction does not spend a significant amount of time
dealing with faults.  If it does, then response time will suffer and the
users will notice.

So I would agree with you that the formula from the manual is more
appropriate than a universal rule of thumb which probably dates from a
previous era.  It might be an interesting experiment to check the
faulting levels as your interactive workload increases.  If you have a
period of time where a few users are active and then usage increases to a
peak time (like 7:00 -> 10:00 on a Monday morning), you could plot
faulting levels and transaction levels during that period.  You mentioned
in your first post that you consistently received a measurement of 4-5
total faults/transaction.  If this number doesn't change as your load
increases, then odds are high that you are not approaching a memory
constraint in your interactive work.  If the number increases with your
load, then you may be approaching a limit within your pool.

Regards,
Andy Nolen-Parkhouse

> Andy,
> 
> I'm not really trying to solve a problem. We are not using the
> performance
> adjuster. All of Qinter runs in one pool - *INTERACT. My questions are
> the
> result of a co-worker who states that if either the DB faults or NonDB
> fault
> values ever goes above 20 then there is a problem with faulting. He's
> seems
> to think the manual isn't relevant. I'm just looking for support that
> there's more to measuring than just looking at the fault rates.
> 
> Gerald