Finding Jobs to Optimize

There are various ways of finding jobs with optimization potential. The most common is of course to listen to users and operation's staff: Are anybody complaining over extensive response or run times, and/or is the server overburdened by long running jobs using extensive resources?

When we at iPerformance look at data collected from a (potential) customer site, we most often are not told which jobs may be of concern. We  normally succeed anyway: GiAPA tells us!

GiAPA’s standard “Job Performance Summary” is a good starting point: By default the sort criteria selected is  5 = CPU time used, but there are as shown in the column to the right many possibilities.

Most commercial applications use more than 2/3 of all resources on I/Os. Therefore we also want to look at jobs appearing in the top of reports sorted on logical or physical I/Os.

Paging is not the problem it used to be, given the much larger memory on the servers today, although the issue pops up from time to time anyway. Sort criteria 15 (Non permanent writes) will show any heavy paging jobs – such jobs could be speeded up if allowed to run in a larger pool, or in a pool without other jobs with higher priority.

For interactive jobs, the report on transaction response time is of course interesting.

The following picture shows a page of the “Job Performance Summary” by CPU usage. GiAPA automatically produces exception reports, not showing the “good behaving” jobs individually. In below example there was 55,556 such batch jobs, which all together had used 68 hours 27 minutes CPU, corresponding to an average of only 4½ seconds per job – we do not need to waste our time looking at such jobs.

(Unless of course there is a small job running thousands of times every day – that could add up to something worthwhile improving – and of course there is a GiAPA report for such cases.)

The analysis of a job is but “a piece of cake”: Just position the cursor on the job, and hit a command key, and the ready-to-use analysis of file accesses or call stacks, or the details per 15 seconds interval will appear.