After the z/OS 3.2 upgrade, some long-running started tasks suddenly
appeared to report much higher CPU usage than expected.
That immediately raised a few questions.
Had the workload suddenly increased?
Was there a performance issue?
Or was the data telling a different story?
Something Didn’t Add Up
Looking more closely at the underlying SMF data revealed an
interesting clue.
Some of our long-running started tasks were not generating
SMF Type 30 Subtype 2 interval records at the expected 30-minute frequency.
Even more strangely, in some cases, data covering multiple
days appeared to be reflected later.
That explained the apparent CPU spike.
The CPU may not have been consumed suddenly. Instead, CPU
usage that should have been reported across several intervals could appear
later, creating the impression of an unexpected spike.
Then came another important clue:
This behaviour appeared after the z/OS 3.2 upgrade.
Mystery Solved
The investigation eventually led to IBM APAR OA69599.
IBM documents a z/OS 3.2 issue where SMF Type 30 Subtype 2 records may not be written for some address spaces. The APAR identifies a problem with the SMF interval processing timer and lists PTF UJ10136 as the fix for the applicable component level.
For more details, see IBM APAR OA69599.
IBM documents a z/OS 3.2 issue where SMF Type 30 Subtype 2 records may not be written for some address spaces. The APAR identifies a problem with the SMF interval processing timer and lists PTF UJ10136 as the fix for the applicable component level.
For more details, see IBM APAR OA69599.