Useful RedPaper: Coupling Facility Performance: A Real World Perspective

(Originally posted 2006-02-24.)

My friend Frank Kyne in the ITSC Poughkeepsie center sent me a link to this draft RedPaper. It provides useful information, from some realistic benchmarks, on a number of Coupling Facility configuration options.

I haven’t exactly been idle in this space recently: One of my favourite clients has sent me their DB2 and RMF data for their production plexes. I won’t comment on my observations (at least not yet). However it led to me building some nice additions to our normal consulting toolset:

1. I slung a view across the RMF view of the structures that support DB2 Group Buffer Pools and the DB2 Statistics Trace view of those pools. The key to the correlation is (fairly obvious) fact that GBP4, for example, has a structure called DSNGRP_GBP4.
2. I repeated this with the LOCK1 structure – DSNGRP_LOCK1 in this example.
3. RMF’s SMF 74 record has a nice field in the Structure section R744QFLG that allows one to decide if the structure in THIS coupling facility is the primary for Duplexing or the secondary. Using this I extended the views in 1 and 2 above to ensure the RMF view was of both the primary and the secondary copies of the structure – where they exist.(This technique can be used for both System-Managed and User-Managed Duplexing structures. LOCK1 is an example of a structure that supports the former and GBPs are the only example I know of for the latter. This client has both. And at least I now don’t look such a twit by not knowing the client is doing Duplexing. And I can tell which CF has the primary copy and which the secondary.
4. In the same vein I can see the different kinds of waits – as in the System-Managed Duplexing case the two CFs use the link between them to synchronise signalling completion to the z/OS image that issued the request. Here’s the piece of grit in the shell: It turns out that the SMF manual incorrectly documents many of the fields in the Request section as being floating point when really they’re 8-byte integers. It took me to dump the wretched records to see why my processing of the data didn’t correspond to what the RMF Postprocessor CF Report said. The documentation will be fixed. Barry? 🙂
5. This client uses shared coupling facility engines (shared ICFs). Last year Mick Owen asked me “when you say CF Utilisation, which one do you mean?” 🙂 So I’ve just had the chance to sling a view across both the RMF LPAR CPU reporting (SMF 70-1) and the coupling facility CPU reporting (fields R744PBSY and R744PWAI in SMF 74-4). It does resolve the issue that Mick was getting at. Some day I’ll blog on what the actual resolution is – or maybe it’ll make a foil in some presentation or other.

So, the art evolves. Well, mine does. 🙂 I have to admit this is the first time I’ve had to ask myself what different XCF Group Names are for. (XCF is recorded in SMF 74-2). So at least in some things I’m still playing catch up.