Re: XCF/GRS question
When we were transforming our environment from separate CPUs/LPARs to sysplex, we did so by subdividing existing systems into sysplex members rather than combining systems into sysplexes. Resulting sysplexes were based on traditional workloads. We ended up with one sysplex that only one member. No other system had the same workload, and no one could justify subdividing it just on principle. No problem. There was one scheduled housekeeping job that did heavy ICF catalog reading. On all sysplexes it ran with x resource utilization except for this one sysplex, where the same job used 2x or 3x resources. I finally asked the question, how is this sysplex different from all other sysplexes? It was also the only parallel sysplex that was still running traditional ring GRS only because with a single system, it didn't seem worth additional CF structure overhead. IBM at the time said for up to four members, GRS ring was adequate. I'm not much into measuring and micro analyzing, so on a hunch I converted this single member sysplex to GRS star. The change was dramatic. Suddenly, with no other changes, the catalog housekeeping job dropped to x resource utilization. This was quite a few years ago. Things may have changed, but I still recommend GRS star for any parallel sysplex regardless of the number of members. . . J.O.Skip Robinson Southern California Edison Company Electric Dragon Team Paddler SHARE MVS Program Co-Manager 323-715-0595 Mobile 626-543-6132 Office ⇐=== NEW robin...@sce.com -Original Message- From: IBM Mainframe Discussion List On Behalf Of Edgington, Jerry Sent: Tuesday, June 2, 2020 10:39 AM To: IBM-MAIN@LISTSERV.UA.EDU Subject: (External):XCF/GRS question CAUTION EXTERNAL EMAIL We are running on single SYSPlex with two LPARs (Prod and Test) with 2 ICFs, all running on the GPs. We are experiencing slowdowns, due to PROC-GRS on Test, PROC-XCFAS on Prod. Weights are 20/20/20/80 for ICF1/ICF2/Test/Prod. We have setup XCF Structures and FCTC for GRS Star Higher Weight: PROC-GRS3.4 users PROC-GRS2.4 users ENQ -ACF2ACB 100.0 % delay LOGONIDS PROC-GRS 99.0 % delay PROC-GRS 13.0 % delay Lower weight: PROC-XCFAS 14.1 users PROC-XCFAS 13.1 users PROC-XCFAS 99.0 % delay PROC-XCFAS 45.0 % delay PROC-XCFAS 16.0 % delay PROC-XCFAS 11.0 % delay PROC-XCFAS 33.0 % delay PROC-XCFAS 77.0 % delay PROC-XCFAS 45.0 % delay GRSCNFxx: GRSDEF MATCHSYS(*) SYNCHRES (YES) GRSQ (CONTENTION) ENQMAXA(25) ENQMAXU(16384) AUTHQLVL(2) RESMIL(5) TOLINT(180) IEASYSxx: GRS=STAR, JOIN GRS STAR GRSCNF=00,GRS INITIALIZATION MEMBER GRSRNL=00,GRS RESOURCE LIST D GRS: RESPONSE=TEST ISG343I 13.38.49 GRS STATUS 604 SYSTEMSTATE SYSTEMSTATE MVSZ CONNECTED TEST CONNECTED GRS STAR MODE INFORMATION LOCK STRUCTURE (ISGLOCK) CONTAINS 1048576 LOCKS. THE CONTENTION NOTIFYING SYSTEM IS TEST SYNCHRES: YES ENQMAXU: 16384 ENQMAXA:25 GRSQ: CONTENTION AUTHQLVL:1 MONITOR:NO Any advice or help on what I can do about these delays, would be great? Thanks, Jerry -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: XCF/GRS question
See below and thank you very much for all the information and suggestions. Jerry From: IBM Mainframe Discussion List on behalf of Peter Bishop Sent: Tuesday, June 2, 2020 8:06 PM To: IBM-MAIN@LISTSERV.UA.EDU Subject: Re: XCF/GRS question This message was sent from an external source outside of Western & Southern's network. Do not click links or open attachments unless you recognize the sender and know the contents are safe. Hi Jerry, questions, and a suggestion. These are more at the hardware layer than the GRS one, which I saw Paul Feller addressing quite well. It may be that you cannot change the LPAR setup, but if you can, here are some ideas. 1. Must the CFs share the GPs with the z/OS systems, or are there ICF engines they can use? - Yes, we don't have CF engines and only 2 GPs, being max'ed out at 16 MSUs For small workloads it may be acceptable to have z/OS and CF workloads in the same processor pool, but CF workloads are different than z/OS ones, and where possible I have seen much benefit from having an ICF pool for CFs, and a CP pool for z/OS (and if you have VM or Linux, an IFL pool, which may be out of scope here). - I am trying to get a CF engine, but I can't get it approved in the budget, yet. z/VM, Linux and IFL are running on a different CEC 2. Must the non-production and production workloads share the same Sysplex? - I started with only one LPAR, running production and Test workloads. Our maintenance window is only once a month for 4 hours. So, we needed a way to "fit" the upgrades into the maintenance windows. I'd be inclined to separate them were I in charge. Two monoplexes may be less hassle than a "forced sharing" Sysplex. But you may have reasons for joining non-production into the production Sysplex. - This is a small system, and I would have a major change to split into separate SYSplex or even two monoplex'es. Due to the way the batch and development are setup, it would be a very big change 3. Do you have DYNDISP=THIN set on the CF LPARs? - Yes, DYNDISP=THIN on the ICF LARs For non-production CFs, this is best, but in your case with a single plex it may be inapplicable. Consider how you might benefit from it. It is a much-improved algorithm than its predecessors has been my experience. Considering you are sharing the pool, it may be a "quick fix" if you can live with it. Try a test. - we two ICF LPARs, so I single SYSplex was really the only way to accomplish all the goals, with minimal impact to the business and developers. 4. If you split the plexes, and have separate CFs, it will be better if you weight the CF LPARs as you do the z/OS ones, e.g. if z/OS has an 80:20 CP pool weight, then the CF LPARs should have the same weights for the ICF pool. - it would be a big deal, both political and business impact to split the environments into more SYSplex'es with more ICF. I think wouldn't be a good idea, with this small setup. - Thanks, I will take a look at the CPU weights, about increasing the ICF LPAR CP pool weights. kind regards, Peter On Tue, 2 Jun 2020 17:39:19 +, Edgington, Jerry wrote: > >We are running on single SYSPlex with two LPARs (Prod and Test) with 2 ICFs, >all running on the GPs. We are experiencing slowdowns, due to PROC-GRS on >Test, PROC-XCFAS on Prod. Weights are 20/20/20/80 for ICF1/ICF2/Test/Prod. >We have setup XCF Structures and FCTC for GRS Star > >Higher Weight: >PROC-GRS3.4 users >PROC-GRS2.4 users >ENQ -ACF2ACB 100.0 % delay LOGONIDS >PROC-GRS 99.0 % delay >PROC-GRS 13.0 % delay > >Lower weight: >PROC-XCFAS 14.1 users >PROC-XCFAS 13.1 users >PROC-XCFAS 99.0 % delay >PROC-XCFAS 45.0 % delay >PROC-XCFAS 16.0 % delay >PROC-XCFAS 11.0 % delay >PROC-XCFAS 33.0 % delay >PROC-XCFAS 77.0 % delay >PROC-XCFAS 45.0 % delay > >GRSCNFxx: >GRSDEF MATCHSYS(*) > SYNCHRES (YES) > GRSQ (CONTENTION) > ENQMAXA(25) > ENQMAXU(16384) > AUTHQLVL(2) > RESMIL(5) > TOLINT(180) > >IEASYSxx: >GRS=STAR, JOIN GRS STAR >GRSCNF=00,GRS INITIALIZATION MEMBER >GRSRNL=00,GRS RESOURCE LIST > >D GRS: >RESPONSE=TEST > ISG343I 13.38.49 GRS STATUS 604 > SYSTEMSTATE SYSTEMSTATE > MVSZ CONNECTED TEST CONNECTED > GRS STAR MODE INFORMATION > LOCK STRUCTURE (ISGLOCK) CONTAINS 1048576 LOCKS. > THE CONTENTION NOTIFYING SYSTEM IS TEST > SYNCHRES: YES > ENQMAXU: 16384 > ENQMAXA:25 > GRSQ: CONTENTION > AUTHQLVL:1 > MONITO
Re: XCF/GRS question
Hi Jerry, questions, and a suggestion. These are more at the hardware layer than the GRS one, which I saw Paul Feller addressing quite well. It may be that you cannot change the LPAR setup, but if you can, here are some ideas. 1. Must the CFs share the GPs with the z/OS systems, or are there ICF engines they can use? For small workloads it may be acceptable to have z/OS and CF workloads in the same processor pool, but CF workloads are different than z/OS ones, and where possible I have seen much benefit from having an ICF pool for CFs, and a CP pool for z/OS (and if you have VM or Linux, an IFL pool, which may be out of scope here). 2. Must the non-production and production workloads share the same Sysplex? I'd be inclined to separate them were I in charge. Two monoplexes may be less hassle than a "forced sharing" Sysplex. But you may have reasons for joining non-production into the production Sysplex. 3. Do you have DYNDISP=THIN set on the CF LPARs? For non-production CFs, this is best, but in your case with a single plex it may be inapplicable. Consider how you might benefit from it. It is a much-improved algorithm than its predecessors has been my experience. Considering you are sharing the pool, it may be a "quick fix" if you can live with it. Try a test. 4. If you split the plexes, and have separate CFs, it will be better if you weight the CF LPARs as you do the z/OS ones, e.g. if z/OS has an 80:20 CP pool weight, then the CF LPARs should have the same weights for the ICF pool. kind regards, Peter On Tue, 2 Jun 2020 17:39:19 +, Edgington, Jerry wrote: > >We are running on single SYSPlex with two LPARs (Prod and Test) with 2 ICFs, >all running on the GPs. We are experiencing slowdowns, due to PROC-GRS on >Test, PROC-XCFAS on Prod. Weights are 20/20/20/80 for ICF1/ICF2/Test/Prod. >We have setup XCF Structures and FCTC for GRS Star > >Higher Weight: >PROC-GRS3.4 users >PROC-GRS2.4 users >ENQ -ACF2ACB 100.0 % delay LOGONIDS >PROC-GRS 99.0 % delay >PROC-GRS 13.0 % delay > >Lower weight: >PROC-XCFAS 14.1 users >PROC-XCFAS 13.1 users >PROC-XCFAS 99.0 % delay >PROC-XCFAS 45.0 % delay >PROC-XCFAS 16.0 % delay >PROC-XCFAS 11.0 % delay >PROC-XCFAS 33.0 % delay >PROC-XCFAS 77.0 % delay >PROC-XCFAS 45.0 % delay > >GRSCNFxx: >GRSDEF MATCHSYS(*) > SYNCHRES (YES) > GRSQ (CONTENTION) > ENQMAXA(25) > ENQMAXU(16384) > AUTHQLVL(2) > RESMIL(5) > TOLINT(180) > >IEASYSxx: >GRS=STAR, JOIN GRS STAR >GRSCNF=00,GRS INITIALIZATION MEMBER >GRSRNL=00,GRS RESOURCE LIST > >D GRS: >RESPONSE=TEST > ISG343I 13.38.49 GRS STATUS 604 > SYSTEMSTATE SYSTEMSTATE > MVSZ CONNECTED TEST CONNECTED > GRS STAR MODE INFORMATION > LOCK STRUCTURE (ISGLOCK) CONTAINS 1048576 LOCKS. > THE CONTENTION NOTIFYING SYSTEM IS TEST > SYNCHRES: YES > ENQMAXU: 16384 > ENQMAXA:25 > GRSQ: CONTENTION > AUTHQLVL:1 > MONITOR:NO > >Any advice or help on what I can do about these delays, would be great? > >Thanks, >Jerry > >-- >For IBM-MAIN subscribe / signoff / archive access instructions, >send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: XCF/GRS question [EXTERNAL]
Jerry, I don't have the type of setup you have but I'll list a few things I might look at. Move the CONTENTION NOTIFYING SYSTEM from your TEST lpar to your MVSZ lpar, assuming MVSZ is your production lpar. This process can add overhead to GRS and on a small lpar it can be noticeable. The process will move around based on which lpar gets IPLed and GRS will not move it back. So we issue the SETGRS CNS=PR03,NP command on our PR03 lpar anytime it is IPLed. Sample of what it looks like on the SYSLOG. SETGRS CNS=PR03,NP ISG364I CONTENTION NOTIFYING SYSTEM MOVED FROM SYSTEM TS03 TO SYSTEM PR03. OPERATOR COMMAND INITIATED. I'm going to guess that the ICF1 and ICF2 lpars may not be getting dispatched as well as you might hope. You might see this in the response times of the CF links or in requests getting switched from synchronous to asynchronous. So that could affect how well they respond to things. Another thing to think about is the size of your IGWLOCK00 structure. The size will be different base on your environment as compared to what I have. I was not the one to size our IGWLOCK00 structure so I'm not much help there. You could try to move some of the XCF traffic off of the CFs and on to CTCs to see if that helps any. Just a guess. Also if the TEST lpar is not getting dispatched when needed that can cause issues for things like GRS and XCF and their ability to respond to request. I hope that some of this is helpful. Good luck. Thanks.. Paul Feller GTS Mainframe Technical Support -Original Message- From: IBM Mainframe Discussion List On Behalf Of Edgington, Jerry Sent: Tuesday, June 02, 2020 12:39 PM To: IBM-MAIN@LISTSERV.UA.EDU Subject: XCF/GRS question [EXTERNAL] We are running on single SYSPlex with two LPARs (Prod and Test) with 2 ICFs, all running on the GPs. We are experiencing slowdowns, due to PROC-GRS on Test, PROC-XCFAS on Prod. Weights are 20/20/20/80 for ICF1/ICF2/Test/Prod. We have setup XCF Structures and FCTC for GRS Star Higher Weight: PROC-GRS3.4 users PROC-GRS2.4 users ENQ -ACF2ACB 100.0 % delay LOGONIDS PROC-GRS 99.0 % delay PROC-GRS 13.0 % delay Lower weight: PROC-XCFAS 14.1 users PROC-XCFAS 13.1 users PROC-XCFAS 99.0 % delay PROC-XCFAS 45.0 % delay PROC-XCFAS 16.0 % delay PROC-XCFAS 11.0 % delay PROC-XCFAS 33.0 % delay PROC-XCFAS 77.0 % delay PROC-XCFAS 45.0 % delay GRSCNFxx: GRSDEF MATCHSYS(*) SYNCHRES (YES) GRSQ (CONTENTION) ENQMAXA(25) ENQMAXU(16384) AUTHQLVL(2) RESMIL(5) TOLINT(180) IEASYSxx: GRS=STAR, JOIN GRS STAR GRSCNF=00,GRS INITIALIZATION MEMBER GRSRNL=00,GRS RESOURCE LIST D GRS: RESPONSE=TEST ISG343I 13.38.49 GRS STATUS 604 SYSTEMSTATE SYSTEMSTATE MVSZ CONNECTED TEST CONNECTED GRS STAR MODE INFORMATION LOCK STRUCTURE (ISGLOCK) CONTAINS 1048576 LOCKS. THE CONTENTION NOTIFYING SYSTEM IS TEST SYNCHRES: YES ENQMAXU: 16384 ENQMAXA:25 GRSQ: CONTENTION AUTHQLVL:1 MONITOR:NO Any advice or help on what I can do about these delays, would be great? Thanks, Jerry -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN -- Please note: This message originated outside your organization. Please use caution when opening links or attachments. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
XCF/GRS question
We are running on single SYSPlex with two LPARs (Prod and Test) with 2 ICFs, all running on the GPs. We are experiencing slowdowns, due to PROC-GRS on Test, PROC-XCFAS on Prod. Weights are 20/20/20/80 for ICF1/ICF2/Test/Prod. We have setup XCF Structures and FCTC for GRS Star Higher Weight: PROC-GRS3.4 users PROC-GRS2.4 users ENQ -ACF2ACB 100.0 % delay LOGONIDS PROC-GRS 99.0 % delay PROC-GRS 13.0 % delay Lower weight: PROC-XCFAS 14.1 users PROC-XCFAS 13.1 users PROC-XCFAS 99.0 % delay PROC-XCFAS 45.0 % delay PROC-XCFAS 16.0 % delay PROC-XCFAS 11.0 % delay PROC-XCFAS 33.0 % delay PROC-XCFAS 77.0 % delay PROC-XCFAS 45.0 % delay GRSCNFxx: GRSDEF MATCHSYS(*) SYNCHRES (YES) GRSQ (CONTENTION) ENQMAXA(25) ENQMAXU(16384) AUTHQLVL(2) RESMIL(5) TOLINT(180) IEASYSxx: GRS=STAR, JOIN GRS STAR GRSCNF=00,GRS INITIALIZATION MEMBER GRSRNL=00,GRS RESOURCE LIST D GRS: RESPONSE=TEST ISG343I 13.38.49 GRS STATUS 604 SYSTEMSTATE SYSTEMSTATE MVSZ CONNECTED TEST CONNECTED GRS STAR MODE INFORMATION LOCK STRUCTURE (ISGLOCK) CONTAINS 1048576 LOCKS. THE CONTENTION NOTIFYING SYSTEM IS TEST SYNCHRES: YES ENQMAXU: 16384 ENQMAXA:25 GRSQ: CONTENTION AUTHQLVL:1 MONITOR:NO Any advice or help on what I can do about these delays, would be great? Thanks, Jerry -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN