(Originally posted 2010-01-16.)
… swear at it. 🙂
No, I KNOW that’s not right – but it’s (for me) an irresistibly bad pun. And it’s a natural reaction, too. 🙂
In a recent customer situation I looked at the RMF Workload Activity Report data for a number of service classes. One WLM Sample count was particularly high: "Capped". In fact I look at, with tooling, SMF and the actual field is R723CCCA. (An IBM Development lab HAD looked at the data through the RMF Postprocessor "prism" and come to the same conclusion.)
It turns out, however, that the service classes in question aren’t part of any WLM Resource Groups. (There IS a service class that is subject to Resource Group capping but it’s not involved here.)
So, how can this be?
A piece of background will help:
The reason I had been asked to look at the SMF data was because a large dump episode had taken rather longer than it should have. It’s the usual lesson of "don’t dump into already busy page packs". The best way to ensure this doesn’t happen is, of course, to dump into memory. (Which might not be affordable, but it IS the best way.)
What had in fact happened was that the system had become under extreme Auxiliary Storage stress. And this had been my suspicion all along.
I’m indebted to Robert Vaupel of WLM Development for confirming this:
Capping delays occur when an address space in the service class is marked non-dispatchable. This can occur when Resource Group capping takes place (switching between non-dispatchable and dispatchable in defined intervals) or when a paging or auxiliary storage shortage occurs and the address space is detected as being the reason for it.
In the above the address spaces are related to dumping, of course.
And the reason I asked Robert was because R723CCCA is populated by a WLM-maintained field (RCAECCAP from IWMRCOLL) – it always paying to understand the source of RMF numbers.
So, if you see values in R723CCCA when Resource Group capping is not in play this might be the cause. I’ve not seen this documented anywhere.
(One thing I’d NOT been crisp about – but Robert firmed up in my mind – is that "Capped" samples have NOTHING to do with Softcapping or LPAR Capping in general. That’s a whole ‘nother story.)
So, there may be a moral tale here: If you THINK the cap doesn’t fit – it might well be the case it doesn’t. 🙂
Mainframe, Performance, Topics 