https://www.researchgate.net/publication/342570694_Coupling_Facility_Configuration_Options_-_Updated_2020
CF is not counted on SCRT, shown on RMF reports. Won't cost you on z/OS, may on some vendors. Thin CFs go to enabled wait when work is completed, restart when interrupt says there is work. Estimate is 3% light sharing to 13% heaving sharing (of z/OS workload). Thin CF would use internal links so no I/O overhead to another CPU. On Fri, Mar 3, 2023 at 9:35 PM Laurence Chiu <[email protected]> wrote: > > The situation. > > We share a couple of Z13's with another (larger client). Z13 B is where we > run our development LPARs and Z13 A is production. > > For critical business reasons an online application on our production LPAR > needs to be highly available and that means in a parallel sysplex. But our > outsourcer has told us it cannot be done for the following reasons because > there are no spare ICF engines on the host B - all are being used by other > CF instances, either to support production Sysplexes or development ones > (not ours). > > Host A does potentially have a spare ICF engine we could use to support a > production parallel Sysplex but good practice does recommend you create a > test one first of course. > > I then asked the question, if host A has a spare ICF engine, can't it be > used to support a CF to be used by the test Sysplex on B. I was advised > this was not possible since there are no spare connections between host A > and Host B (Infiniband possibly) so the Sysplex on B could not actually > communicate with the CF on A. > > Our requirement for the Sysplex is primarily to be able to share a VSAM > dataset which is hit every time a transaction comes in with a peak of about > 99tps. So we would need VSAM RLS to share the dataset records between the > two application instances. There is no DB2, CICS or IMS so I think the only > structures in the CF are those to support VSAM RLS, maybe some XCF > structures and core systems. > > Knowing that we would only bring up the test sysplex to make sure > transactions routed correctly across the two LPARs and most of the time we > would have one member of the Sysplex off, I suggested that the test CF > could be built using a CP. To this suggestion I received the following > (anti) advice > - there would be MSU costs (we don't care since we think the MIPS load on > the CF would be low). Plus we would ask that the CF be defined with Dynamic > Coupling Facility Dispatch and set DYNDISP=THIN. Since that CF is going to > be idling most of the time, MSU consumption is not going to be a major cost. > - it's strongly recommended not to do this by IBM. Yet when I read this > document > > https://www.ibm.com/downloads/cas/JZB2E38Q > the option is discussed in great detail and the only negatives are the > incurring of MSU costs and some performance degradation if both a z/OS and > CF LPAR are trying to use the same CP at the same time. But this can be > managed. > > - that a CF running on a CP would need a dedicated CP engine and there are > no spare engines in host B. That totally flies against the information I > have read from IBM docs. > > Of course for production the CF on host A would be configured to use an ICF > engine (or share one) > > Finally, while I accepted the argument at the time there were no > connections between Host A and Host B, further reading suggests that you do > not need to dedicate channels for communications but use XCF or by using > Infiniband sub channels or sharing the same physical link with more than > one Sysplex. Then the issue of running the CF on a CP goes away since I can > ask for two CF's to be defined on host A, one for production and one for > test and DCFC ensures that that production CF is not impacted by the > development one. > > A lot to digest here but I really want to have some authoritative data in > order to refute most of the comments being our outsourcer. > > Thanks > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to [email protected] with the message: INFO IBM-MAIN -- Mike A Schwab, Springfield IL USA Where do Forest Rangers go to get away from it all? ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
