Re: Got a weird suggestion about a troublesome Tomcat installation running in a subsystem

On 2/2/2017 7:16 PM, James H. H. Lampert wrote:

On 2/2/17, 12:02 PM, Buck Calabro wrote:

When you measured the performance of the machine, did anything jump out?

What sort of performance metrics did you have in mind, that would be
specific to this situation?

WRKSYSSTS. Collect between 5 and 10 minutes of data. F11 through all
the displays. The red flags for me are:
Anything to Inel
High faulting rates
High CPU%

WRKDSKSTS. Collect between 5 and 10 minutes of data. F11 through both
displays. Are any units out of line with the rest?

WRKACTJOB CPUPCTLMT(5) Does anything show up in 5 minutes or so?

GO PERFORM has lots of tools.

Unfortunately, what I know about performance tools and metrics,

There is an entire section of the Knowledge Centre dedicated to
performance:
http://www.ibm.com/support/knowledgecenter/ssw_ibm_i_73/rzahx/rzahx1.htm
Unfortunately, few of us needed to look at this, because the machine is
magnificent by default. So we (and I count myself in this!) tend to
jump straight into 'Researching a performance problems'
http://www.ibm.com/support/knowledgecenter/ssw_ibm_i_73/rzahx/rzahxresearchparent.htm


What you're really looking for are deviations from the norm. And yes, I
fully comprehend there's no norm to compare against; I've been there
too. Until we/someone measures the system often enough to have a
history, all we can do is triage and try to spot the reason.

We've switched them to a version of the subsystem without the private
memory pool; response time hasn't changed significantly.

I personally have never had any success with performance tuning by using
my instinct to try this or try that. The only time I have had success
tuning a machine is:
1) Measure
2) Analyse
3) Change One Thing
4) Record what changed
5) Iterate