Re: WRKSYSTS Page Fault Values
- From: "GB" <gb@xxxxxxxxxxxxx>
- Date: Sun, 16 Jul 2006 04:55:15 GMT
Hello, fellow iSeries professionals. In an attempt to gather some
basic performance information for our EOD processing window, I
scheduled a CL program that does the following:
Firstly a general comment, 5 minutes is too short a snapshot to tell you
anything meaningful!
1. Do the +++++'s in the page fault columns mean that faulting was so
bad that the values were too high to display? Or could this be due to
the fact that the subsystem was ended and restarted during the elapsed
time?
Not sure on this one. 29.29mb isn't a very big pool is it? Try running your
little capture again (over say 15 minutes) and check again.
BTW what runs in the ATM subsystem?
2. Wait-Inel and Act-Inel are 0.0 across the board. Should I be aiming
for this and thus leave it alone? Should I lower my activity levels in
hopes of reducing the fault rates?
0.0 is v. good, there are non-zero ratios for these number that are also
acceptable. 0.0 means system is cruising...
3. System value QPFRADJ is set to 2 (Adjust at IPL and automatic). Is
the overhead of automatic adjustment really bad?
Not at all. Overhead is v. low - you will always get more back in efficiency
than you would lose...
4. I was thinking about turning off automatic adjustment and using
CHGSHRPOOL/CHGSBSD twice a day to toggle between daytime and evening
processing settings. Is this a good idea?
Daytime & Nightime/EoD configs do work - but if the auto adjust is on, it
will retune the machine in a matter of 2-5 minutes anyway.
5. Are my disk arms severely out of balance? Unit 53 seems to be
getting many more IO requests than the others but it's a different
story during the day - the balance is even worse but for different disk
units.
This is a 5 minute snapshot. A single program reading a large file from
beginning to the end can push a single disk to those sort of levels. Long
term averages are what matter. 5 minutes isn't enough!
Can I improve things with STRASPBAL and/or STRDSKRGZ?
They can't do any harm. They can be stopped and started and also set to run
for a period of time before stopping. You can therefore schedule a couple of
hours a night for a week or whatever. It won't do any harm - but your disk
utilisation (amount full) is very consistent at 69.3/69.4 so I don't expect
that it will make any difference to be honest.
6. Should I be happy with the low "% Busy" values and not worry about theIO requests?
Yes, I/O high requests is generally a measure of throughput - you do want
your system to do some work don't you?
7. I have 7 days worth of Performance Data. What should I look at next?
Do you have the "Performance Tools" software? There are oodles of reports -
but you need to have a problem to solve first. On the basis of the 5
minutes, I can see nothing wrong with the machine - is there a performance
problem that you have?
26% on the CPU and <3% on the disks means that the system was virtually on
tickover...
HTH
GB
.
- Follow-Ups:
- Re: WRKSYSTS Page Fault Values
- From: Madcap
- Re: WRKSYSTS Page Fault Values
- References:
- WRKSYSTS Page Fault Values
- From: Madcap
- WRKSYSTS Page Fault Values
- Prev by Date: Re: WRKSYSTS Page Fault Values
- Next by Date: Re: use QPRCRTPG API
- Previous by thread: Re: WRKSYSTS Page Fault Values
- Next by thread: Re: WRKSYSTS Page Fault Values
- Index(es):
Relevant Pages
|
