I've done some digging, and now I need to ask how we got to where we have things set up. Here's how we're configured: The machine is a z16-AGZ M02 with two CPs and two ZIIPs configured. We have three LPARs: production, text, and sandbox/installation. All three LPARs' configuration pages have 0 online and 2 reserved zIIPs (and 2 online and 1 reserved CP). In the production and test LPARs, COMMND00 has 'CF CPU(3),ON' and 'CF CPU(4),ON' commands, which configure on the two zIIPs. And the offending IPL had an IEA675I message about two transient zIIPs used for boost.
I have no idea why we configured the profiles with 0 online zIIPs and then explicitly configure them online in COMMND00. My guess as to what happened is that the processors were configured offline at IPL time (we got IEE496I messages about that), and the boost activation happened before COMMND00 was processed, and so when the boost ended, it didn't know that we'd explicitly configured those online and happily configured them off. Seems to me that the easiest answer is to rip out those two commands in COMMND00 and change the activation profile to have two zIIPs online to the test and production LPARs. (We normally do not run with any online to the sandbox.) Is there a reason why we might not want to do it that way, or for that matter, why we might have wanted to do it as we are now to begin with? It's entirely possible that that was an artifact of our migration from a z10 to the z14 several years ago. Don't you just love systems archaeology? On Wed, Dec 17, 2025 at 1:49 PM Paul Feller < [email protected]> wrote: > Jay, I have to ask. Are the zIIPs set in the lpar profile to be online at > IPL time? By chance are you normally bringing the zIIPs online after the > IPL is done? > > If I remember correctly the boost process by default will only keep online > the zIIPs it see as being online at IPL time. That would be those zIIPs > you > have configured in the lpar profile to be online. > > So, if the zIIP setting for INITIAL is set to 0 and RESERVED is set to 2 > after the boost process is done it would take the zIIPs offline. On the > flip side from what I remember if INITIAL is set to 2 and RESERVED is set > to > 2 then when boost is done there should be 2 zIIP online. > > I don’t recall ever using the PRESCPU option. > > From the MVS Initialization and Tuning Reference > PRESCPU > This parameter causes system initialization CPU processing to bring > logically online those CPUs (and only those CPUs) that are physically > online > when the IPL is initiated, without regard to the number of CPUs defined to > be initially online in the logical partition profile. > > The purpose is to ensure that all CPUs that are online when an IPL is > initiated are online when the IPL completes. > > CPUs that are offline when an IPL is initiated remains offline. You might > not want to use PRESCPU when WLM CPU management is active because that > processing might be configured CPUs offline. > > If PRESCPU is not coded, CPU initialization takes CPUs offline or brings > CPUs online as needed to make the number of online CPUs equal to the number > of initial CPUs defined for the partition, when that information is > provided > by the machine (as it should be on model 2064 (GA-3), model 2066, and later > machines) and is not 0. > > If the number of initial CPUs is not provided or is 0, CPU initialization > attempts to bring online all CPUs that are physically online. On most > models, those are the CPUs that were online just before the IPL was > initiated. However, models 2064 and 2066 can configure additional CPUs > online when an IPL is initiated, and CPU initialization also brings those > CPUs online. in System Recovery Boost. > > For information about how PRESCPU interacts with BOOST, see Interaction of > Shutdown Boost an > > > > Paul > > -----Original Message----- > From: IBM Mainframe Discussion List <[email protected]> On Behalf > Of > Jay Maynard > Sent: Wednesday, December 17, 2025 10:07 AM > To: [email protected] > Subject: Nasty IPL Boost surprise > > We cut our production LPAR over from a z14 to a z16 this morning, yay! We > did discover one thing the hard way, and it came as a nasty shock. > > The z16's IPL Boost feature is a nice thing. It basically unleashes the two > zIIPs we have configured (alongside the two CPs on a z16-M02) to do CPU > workloads as well. It did make our system come up much quicker. But when > the > hour was up, the system ended the boost by configuring the two zIIPs > completely offline. We didn't discover this until almost three hours later > when we realized that no work was getting dispatched on the zIIPs. We > recovered by configuring the two zIIPs online. > > We're an Adabas and Natural shop, and essentially our entire application > workload is zIIP eligible and normally runs that way. Not having the zIIPs > running would have severely impacted our system's performance. > > Is there a way to have the system end boost not by taking the zIIPs > completely offline, but making them zIIPs again? Or is the best approach > here to automate a response to the IEF678I or IWM064I messages to issue the > CF CPU command? > -- > Jay Maynard > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, send email > to [email protected] <mailto:[email protected]> with the > message: INFO IBM-MAIN > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to [email protected] with the message: INFO IBM-MAIN > -- Jay Maynard ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
