Re: RSH Consulting - RACF Survey - June 2019 - Performance - ERV
Prior to to z/OS 1.3 (i.e., some 15 years ago) there were reasons for increasing the value of the ERV parameter and from time to time one can still see values of up to 50,000. Since the ERV is being restarted there are no good reasons for such exaggerated values and I'd consider the default of 500 to be good. And just to extend the topic a bit - the IEAOPT parameters that one would frequently want to specify non-default values for are: -CPENABLE=(10,30) -ManageNonEnclaveWork=Yes (when using WAS/Liberty with enclave management) -BLWLIntHD=5 for more aggressive blocked workload support, namely for DB2 And for both RCCFXET and RCCFXTT AUTO should be in place. Then there are a few that deserve consideration, such as INITIMP=E, MT_ZIIP_MODE=2. It is possible that the defaults will eventually change:-) Horst Sinram - STSM, IBM z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: honorpriority=no in WLM
The OP's question was about DB2 workloads. Resource group capping for DB2 workloads would be pretty risky unless you could really guarantee that you do not share resources with your production work. An RFE for period level resource groups has been rejected in the past for both technical reasons and and the (significant) effort. And, to add some larger context: For those installations that decide for themselves that the consumption based Tailored Fit Pricing model is the right way to go the whole capping discussion will become a moot point. Horst Sinram - STSM, z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: honorpriority=no in WLM
Duncan, why do you believe that HonorPriority=No could help? You will only be confining work to the already potentially overloaded zIIPs by disallowing help from the CPs. So, HonorPriority=No can only make things worse. I assume that you don't want to go so far and configure all zIIPs offline to that partition; the way to go is really having sufficient zIIP capacity for your workload peaks. Horst Sinram - IBM z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: SAP Processor Utilization
On z/OS, you can use SMF78.3 and the RMF IOQ report. Plus, there are some related overview conditions. Horst Sinram - STSM, IBM z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CBU expiration and activate dynamic
Multiple CBU records may be active at the same time. CBU resources can be activated, deactivated, or the activation level changed concurrently (from an LPAR perspective). I assume you're using shared processors only and z/OS. Usually, from z/OS you may need to configure additional logical processors online to utilize more logicals from a given LPAR, or (recommended) configure logicals offline before physically deactivating resources, if the new number of physical processors in the shared processor pool would be less than the number of logicals online to any single partition. You cannot extend the CBU resources *from a given record* beyond the grace period. The Capacity on Demand user's Guide for your hardware model is a good resource, e.g., regarding the grace period handling. In general, there should be no reason for IPLing z/OS unless you had disabled DYNCPADD in your LOADxx, or you would be running out of the limit and you need to utilize more logical processors. z/OS subsystems are usually well prepared to deal with model changes. Horst Sinram - STSM, z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: "Workload" field in SDSF
Hi Andy, there are many different ways to obtain the workload name. The probably simplest is to issue a Sysevent REQ(F)ASD against the space and find the workload name in the IRARASD as RASDWKLD. Horst Sinram - STSM, z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Novice query on WLM velocity
Peter, classification is under control of the classification rules in your WLM service definition. For started task, you need to check the classification rules in the STC subsystem. For STC a few particularities apply: (1) The default service class, if not assigned differently, is SYSSTC (2) System-defined properties (aka "SPM rules"), such as from SCHEDxx, are automatically into the classification rules. If you do not explicitly include SPM SYSTEM and SPM SYSSTC they will be pulled in logically behind your own classification rules. The SPM rules can classify into the SYSTEM or SYSSTC service classes (3) Some system address spaces will always be classified into SYSTEM, or at least SYSSTC, for reliability reasons. https://www.redbooks.ibm.com/abstracts/sg246472.html?Open is a bit older but probably still a good place to get started. All the glory details are in https://www.ibm.com/support/knowledgecenter/SSLTBW_2.3.0/com.ibm.zos.v2r3.ieaw100/toc.htm Horst Sinram - STSM, z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Service class changes
See SMF Type 90, subtype 30: https://www.ibm.com/support/knowledgecenter/en/SSLTBW_2.1.0/com.ibm.zos.v2r1.ieag200/iea3g2_RESET_command_complete.htm Horst Sinram - STSM, z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Measuring Usage of zIIP per address space
The SMF30 Processor Accounting Section is a good place for the total and zIIP processor consumption but if you are also interested in the delay aspects you won't find anything in SMF30. For delay information you can turn to RMF Mon III, or just use the postprocessor reports with report classes defined over those address spaces that you're interested in. As to what fields you should look at, be sure to read to the introduction to SMF30 in the SMF manual. If enclaves are involved then https://www.ibm.com/support/knowledgecenter/en/SSLTBW_2.3.0/com.ibm.zos.v2r3.ieaw200/ieaw20046.htm describes the accounting scheme. Horst Sinram - STSM, IBM z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Where is the WLM website???
Hi Kees, glad that someone notices :-) But seriously - while some restructuring of the web site is ongoing we're working on cutting over to mostly github based repositories. If you open the presentation at https://github.com/IBM/IBM-Z-zOS/blob/master/zOS-WLM/WLM%20Tools.pdf you'll see a summary of the WLM related tools. Alain's LPAR design tool can be downloaded from ftp://public.dhe.ibm.com/eserver/zseries/zos/wlm/LPARDesign-HD-zPCR-V9-T03_IBM.zip We expect a restructured "web page" to become available shortly at https://github.com/IBM/IBM-Z-zOS/tree/master/zOS-WLM/WLM%20Documents.md (It is NOT up and running as of today). Horst Sinram - STSM, IBM z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Can a job determine its own WLM priority?
Phil, there is no "WLM priority". There is a dispatch priority, and there is a WLM importance (and goal). - Which one are you interested in? - Only for simple cases the entire "job" (address space) would be managed to the same service class period, and therefore dispatch priority. Are you definitively not interested in e.g. in enclave management? - Is the "job" authorized? Horst Sinram - STSM, IBM z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: WLM?
Anne, a daemon is a long running task, like an STC. It makes very little sense to use multi period goals (with short durations) because it will eventually fall through to later periods. A single period velocity goal is the preferred goal type for a daemon. Only if the daemon would be at risk to loop you would consider a multi period goal using a *really huge* duration value. Even in that latter case a resource group would be a better safety net. Horst Sinram - STSM, IBM z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Advice for WLM tuning of OMVS forked address spaces
Hi Kirk, my initial assumption would be that the "spike" effect they're seeing is not specific to the fact that OMVS initiators are being used. As you mentioned,it is important that the work that newly arrives into the system is properly classified and has not overly aggressive goals. However, it is equally important that the other "important" is correctly classified with a goal that is aggressive enough to protect against other work. This last item is "usually" what contributes most to such unwanted effects. Then, depending on the JCL being used jobs may consume some (or more) CPU resources during initiation before classification is in effect. For such cases there is the IEAOPTxx INITIMP parameter. Specifying INITIMP(E), or one of the numeric values, depending on the workload to be protected can be useful to minimize the spike effect due to incoming work Horst Sinram - STSM, z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Hiding cross LPAR capacity information
In addition, by removing Global Performance Data Control authority one would also disable some CPC-wide optimizations in z/OS HiperDispatch. Horst Sinram - STSM, IBM z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: DB2 Locks and WLM Blocked Workload Support?
Peter, DB2 has the most comprehensive support to interact with WLM to address contention. DB2 will try to notify WLM about resource contention to extent it knows about dependencies. Based on such notifications DB2 and WLM support - Regular "enqueue promotion" and short term promotion . Both promote the holder of the lock to an elevated dispatch priority, increasing its chances to get dispatched. Search for "Sysevent ENQHOLD") - Chronic contention. Elevates the resource holder to the highest dispatch priority of any waiter. (Search for IWMCNTN). Still there can be situations where neither DB2 (and certainly no one else) knows the dependencies. For such cases the blocked workload support provides the capability that *any* address space that has been blocked -not dispatched- for a given time will be granted one time slice. That's a very small amount of processor time and it is handed out *independently* of any contention. Yet it is proven to be a very effective way to address DB2 latch contention. (Only latches typically gate such a short path that the contention can be resolved through one or a few of such "trickles".) See the ENV and BLWLINTHD IEAOPTxx parameter (BLWLTRPCT is almost never the limiting fact). Searching for OA44526 will give you a good list of recommendations. DB2 and GRS locks are unrelated (except for allocation related ones). Horst Sinram - STSM, IBM z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: LPAR is 100% capped but MVS is busy only 60%. Not too bad, right?
Peter,
> When an LPAR is capped due to the R4HA being exceeded, RMF 3 CPC report shows
> this as "WLM Capping %:" being greater that zero. But this alone is
> not an indication if the capping really hurts. If MVS busy (RMF 3 SI report)
> is nowhere near 100%, then the situation is not really bad.
As you already mentioned, the %WLM Capping metric just indicates to what extent
during the interval a cap had been in place. 100% just means that phantom
weight capping was used. The more relevant metric is the %WLM Actual (or %ACT)
metric that identiefies to what extent the consumption of the LPAR was close to
the effective limit.
>Say WLM capping is 100% for a 300 second period, and MVS busy is 63% for the
>same period, then I would say everything that wanted access
>to processors should have gotten access.
Over the interval(!) MVS did still decided to go into a wait 37% of the time.
It may still worth while to check %Actual capping as well as the In-Ready work
unit queue distribution (or processor delays in the Workload Activity Report).
>However, those task working on the processors will work slower due to PR/SM
>taking away the physical CPs from the logicals due to capping being active.
Less dispatch time on the physical processors will be available.
>This is true even for vertial high (pseudo dedicated) CPs.
More precisely: If the LPAR is capped to an MSU limit below it's weight
equivalent (positive phantom weight capping) it's PR/SM entitlement (weight)
gets reduced. That means that vertical highs could become mediums or even lows.
But those logicals that remain VHs will continue to be dispatched as VH
("pseudo dedicated")
In
ftp://public.dhe.ibm.com/eserver/zseries/zos/wlm/Capping_Technologies_and_4HRA_Optimization_2016.pdf
you'll find some more details.
Horst Sinram - STSM, z/OS Workload and Capacity Management
--
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Would HiperDispatch likely delay heavy multitasking job?
Peter, your second bullet isn't really correct. First, as described in my previous post, nodes will be able to help each other. Then, when deemed required, TCBs may also be broken out of their mother address space and assigned a different affinity node. Lastly, the process is independent of goal achievement. Therefore, it is correct to say that for increased efficiency, HiperDispatch strives at dispatching related work close (on a node). However, as you could quickly proof, a heavily multitasking address space isn't necessarily confined to a single node. Kind regards / Mit freundlichen Gruessen Horst Sinram - STSM, IBM z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: AW: Re: Would HiperDispatch likely delay heavy multitasking job?
Peter, also within the same processor class (here CP) affinity nodes may help each other - see the IEAOPTxx CCCAWMT parameter. Therefore, the number of (unparked) logicals processors on a given affinity node does not necessarily limit the number of concurrently disptached work units of an address space. Giving the work unit a high priority can help increasing the parallelism because the helper node would select by priority. Horst Sinram - STSM, z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: WLM Question
> Is there a way in batch to extract a WLM service definition and perform a > save/save as to a dataset that I can use for DR preprocessing purposes? Bob, depending your further requirements there are two RFEs (requirements) that you could consider voting for https://www.ibm.com/developerworks/rfe/execute?use_case=viewRfe_ID=57448 which is pretty much along the lines of what you're asking for. Unfortunately it wasn't created as public, so probably need to create another one (using 'public' is always a good idea). Then http://www.ibm.com/developerworks/rfe/execute?use_case=viewRfe_ID=73305 includes the capability of writing to a not active CDS. Of course you could begin with a programming exercise (IWMDEXTR was already mentioned) but you're looking for what could be done with existing means I'd recommend that you - converge (as everyone should...) to an XML format (sequential) service definition (i.e either use z/OSMF or 'save as' XML in the ISPF application - Use naming conventions to allow you determine the latest level of the service definition (yes, enforcing conventions isn't bullet proof but probably good enough) - Include the latest service definition(s) on your DR backup sets Horst Sinram - STSM, z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: can a program determine the capacity setting of a z-box?
Sysevent REQLPDAT is z/OS only. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: can a program determine the capacity setting of a z-box?
Sysevent REQLPDAT offers quite a bit more information than Sysevent QVS (https://www.ibm.com/support/knowledgecenter/SSLTBW_2.2.0/com.ibm.zos.v2r2.iead200/iead200821.htm) and could also be called unauthorized when that matters. Horst Sinram - STSM, z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Silly problem with WLM Classification Rules
Hi Kees, sorry, there is no way to classify based on the existence/specification of a work qualifier. On z/OS V2.1 you could simplify your rules by creating a scheduling environment group but it would still require that the group specification includes all your scheduling environments. Horst Sinram - z/OS Workload and Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: ENF 55(SRM Shortage) event
Miklos, it depends on the qualifier what variables are filled in. We intend to add commentary to provide that information. In the meantime, SYS1.SAMPLIB(IRAEN55S) has a sample exit that you may find useful. Kind regards / Mit freundlichen Gruessen Horst Sinram - IBM z/OS Workload Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Is it possible to open PCOMM session up to 50?
Personal Communications Version 6.0 supports up to 52 sessions. See http://pic.dhe.ibm.com/infocenter/pcomhelp/v6r0/index.jsp?topic=%2Fcom.ibm.pcomm.doc%2Freadme%2FreadV60.html I think you need to use sessions A-Z and a-z -- but I never tried myself :-) Horst Sinram - IBM z/OS Workload Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: why does WLM Server status change from YES to NO
Any region being managed to transaction goals has *always* been managed to both - startup and shutdown use the region goal. I don't understand why this option was introduced - and why it is only recommended for TORs. Guess I'd better go find some doco (pointers gratefully accepted). Shane, you could refer to ftp://public.dhe.ibm.com/eserver/zseries/zos/wlm/WLM2012Share.pdf (or navigate there via http://www.ibm.com/systems/z/os/zos/features/wlm/). As described there Mananage to Both is intended to combine the region management of the TORs (allowing to favor the TORs over the AORs) with the full transaction reporting that you would get from Manage to Goal of Transactions. It's primarily useful for systems where CICS is the predominant workload. (And, yes, I have seen data from quite a few large sites using it very successfully.) Horst Sinram - IBM z/OS Workload Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: why does WLM Server status change from YES to NO
we have a single OLTP region, which is TOR and AOR together. Would you suggest managing the region using goal of transaction or both ? Walter, in that scenario you would probably stick to TRANSACTION if you're satisfied with the way that region is currently managed by the transactions running in it. Horst Sinram - IBM z/OS Workload Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Defined capacity
It is probably true, that during the uncapped parts of the pattern, IRD could have a chance to adjust weights, if this is what you mean by a 'better chance'. Yes, exactly. (BTW, I don't want to promote pattern capping in general.) AFAIK, I must use the HMC to set the values again to their initial values (or do a POR ;-). Altogether, I did not dare to activate IRD after activating GCLs. At the HMC, you could set MIN=MAX=INITIAL weight to enforce the weight setting that you want to implement and transition then. Also, there are APIs that allow driving such changes automatically. These APIs could be used through some products (e.g. Tivoli System Automation ProcOps), or you could (probably, but I did not verify that all operations are supported e.g. in z/OS BCPii) come up with homegrown automation. Horst Sinram - IBM z/OS Workload Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: WLM Intelligent Resource Director (was Defined capacity)
Now we know, that Hiperdispatch also disables fundamental IRD functionality. That statement is not correct: - HiperDispatch *replaces* Vary CPU Management (and only when HiperDispatch=YES.) - HiperDispatch provides *more* functionality and is much more efficient and faster than the old Vary CPU Management such that there is no reason why one would want to continue with the old management in a HD=YES environment Horst Sinram - z/OS Workload Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Defined capacity
This at least suggests that IRD Weight Management is also disabled when an Lpar is softcapped by DC limits and not only by GC limits. Kees: Correct - when an LPAR is being capped, regardless whether that's due to an LPAR level defined capacity, or due to group capacity, IRD won't help the partition. In the case of LPAR level DC that's likely no problem: The LPAR's 4-hour-rolling-average has exceeded the installation defined limit despite the fact that the LPAR already has a low weight. Granting it even more weight doesn't make much sense. With group capping the situation is a bit different: the LPAR weight does also determine the LPAR's entitlement of the group capacity. With weight being the current (vs. initial) weight IRD may have contributed to managing the LPAR to a low entitlement. That can sometimes be problematic, e.g. depending on whether the importance distribution of your workload within the LPAR cluster has changed or not, and how long the capping situation persists (with pattern capping there is obviously a much better chance.) Horst Sinram - IBM z/OS Workload Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Low priority workload
Anyway, I decided to define a resource group with just 100 SUs to force batch down. Surprisingly batch still used up to 2000 SUs, because WLM promoted the batch workload due to any blockings, enqueues or locks batch held. So promotion by WLM might be another reason at your site that batch runs better than expected. You can verify this with the Workload Activity report, it includes a column Service and Promoted. Hmm, did you verify that 2000 SU/sec were in fact used at a promoted dispatch priority? A z196 model 7xx (just as a typical example) delivers between 33,000 and 61,000 SU/sec per processor. Resource groups work by marking the work dispatchable/non-dispatchable for multiples of 1/64th of the time. Therefore, a single logical CP could deliver between roughly 500 and 950 SU/sec. Depending on the type of resource group, the number of logical processors within the scope of your resource group (i.e. system or Sysplex), and the amount of work running at a higher priority you may well end up at that order of magnitude for the achievable granularity for *that* resource group in your environment. Horst Sinram - IBM z/OS Workload Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: ''MVS overhead'' (as indicated by OMEGAMON)
Jan, it appears the monitor is trying to tell you that you have a low capture ratio. Searching for that term should give you a good understanding of what the message really means, and in a lot of good advice, e.g. in the Effective zSeries Performance Monitoring Using Resource Measurement Facility publication. (IMHO the term overhead is misleading, 'management time' would be more accurate). If you believe there might be a problem you may want to start with verifying LPAR configuration and IEAOPTxx settings (any obscure timing parameters? Anything violating recommendations?). You could also verify that the capture ratio is indeed low using SMF70/72 data. Phantom weight is irrelevant in this context. It is only a vehicle to tell PR/SM how to cap the partition. The fact that the LPAR *is* capped at that time (to what extent?) may very well be relevant, though: More work, longer work queues... CPUMF counters won't help you diagnosing anything. CPUMF sampling could help -theoretically- but the data may be very hard to evaluate. Horst Sinram - IBM z/OS Workload Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: SMF Type 99 quandry
According to the MVS Programming Workload Management Services manual, *a* Type SMF 99 record is written every policy interval (approximately 10seconds). Mark, I don't think the book is saying that. It says SRM writes type 99 record*s* for each policy interval, or approximately once every 10 seconds. In the MVS System Management Facilities (SMF) book (http://publibz.boulder.ibm.com/epubs/pdf/iea2g2c1.pdf) you can find a summary of what subtypes are written. And subtype 12-14 are written every 2 sec but they are very small as well. Kind regards - Horst Sinram IBM z/OS Workload Management Capacity Management -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
