|
PERFORMANCE ANALYSIS

Having
data available for all jobs every 15 seconds, the
possibilities for different reports and charts are almost
infinite. This report is often chosen as the first to look
at the results from an analysis: “Job Performance Summary”
descending by CPU time used.

Here are a
few examples of results and options based on above data – we
refer to the red numbers:
1.
Batch job SNBRVRE consumed for hours almost an entire CPU –
715 million reads and only 1613
disk
accesses. Cause was easy to spot: Job was looping!
2.
19066 batch jobs “behaved” so well that GiAPA decided not to
keep them individually for detailed
analysis – but all details are available, just in case…
3.
An interactive job should never use more than one hour CPU –
the user probably ran batch-type
programs! Placing the cursor on the job and hitting F6, F7
or F8 will give analysis of files accessed
(+
spot inefficient I/Os), and F9 will run call stack analysis
(tell which program(s) used most
resources)
F10 / F11 give further details.
4.
To see graphical representation of e.g. jobs using more than
½ hour CPU, place the cursor on the
last job above the dotted line and hit F14. The chart data
will contain CPU seconds used, because
the
report sequence was descending on CPU usage.
By default
a bar chart is selected:

But of
course the user may select between different types

As stated
above, there are so many possibilities. Here is a 3D column
chart showing the users causing the highest number of disk
accesses.

This
report shows which user programs that most often caused a
HotSpot, i.e. were active when a job had used more than 6 %
CPU the last 15 seconds. Optimizing the top-10 here is
likely to cure 80+ % of all performance concerns … (F14
would have made the data available for graphic display.)

If you
know that there was a performance drop at a given point of
time, you can of course get statistics per 15-seconds data
collection interval. It also shows the system ASP disk usage
percent, and the names of the 6 jobs that used the most
resources at each interval.

BACK TO TOP
|