Re: CPU Utilization
Hello, I am able to get RMF3B messages on on my syslog. But I unable to find the way to get these message like below when DASD activity rate goes above limit on console. Can you please suggest way of achieving this task . +RMF100I 3B: Processing WFEX Report... +RMF100I 3B: Name Reason Critical val. +RMF100I 3B: -- +RMF102I 3B: *STOR ONLXF < 16K 0 frames +RMF102I 3B: OSPCAT DAR > 5 9.000 /sec +RMF102I 3B: OSPRES DAR > 5 7.300 /sec -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
Many years ago when CMOS machines (9672) first came on the scene, individual CPs were a lot slower--by 3x--than the bipolar CPs they were replacing. Not a big issue for transactional processes like CICS, but batch suffered because of long standing jobs that suddenly were getting S322 abends doing the same amount of work as before. Rather than force JCL changes for hundreds of existing jobs, we added code in IEFUTL to extend the time for two additional intervals of whatever TIME= value was in effect. We kept track of extensions via some flags and issued a message each time the limit was extended. Eventually of course CMOS speeds caught up with bipolar and exceeded it. Not surprisingly, our IEFUTL still grants extensions. If it ain't actually broke... . . 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 [mailto:IBM-MAIN@LISTSERV.UA.EDU] On Behalf Of Joel C. Ewing Sent: Wednesday, July 04, 2018 10:07 AM To: IBM-MAIN@LISTSERV.UA.EDU Subject: (External):Re: CPU Utilization When total CPU time used by the address space is the issue, that can be handled using an IEFUTL exit, setting a default JOB TIME parameter by job class in JES2, and overriding that with TIME parameters on the JOB and EXEC statements as appropriate. The IEFUTL exit can be written to either cancel the job when the limit is exceeded, or ask the operator whether he wants to extend the CPU TIME limit or cancel the job, and differentiate based on Job class, etc. I suspect you could probably write the IEFUTL exit to auto extend the limit for an STC address space, but put out a console message so the Operator would know it was happening, and he could then manually cancel the address space if the behavior and CPU extension was unreasonable. Joel C. Ewing . On 07/04/2018 12:31 AM, saurabh khandelwal wrote: > I would like operator console to be notified when any address space or > Job taking more CPU for longer then any time we specify > > On Wed, Jul 4, 2018 at 8:27 AM, retired mainframer > > wrote: >>> -Original Message- >>> From: IBM Mainframe Discussion List On >>> Behalf Of Peter Hunkeler >>> Sent: Tuesday, July 03, 2018 10:10 PM >>> To: IBM-MAIN@LISTSERV.UA.EDU >>> Subject: AW: Re: CPU Utilization >>> >>>> No, IEFUTL is called INSIDE an Address Space as often as specified >>>> in >> SMFPRMxx in >>> statement JWT. >>> >>> >>> and it would also be called when the CPU time used exceeds the >>> time >> limit of the job or >>> step (TIME= parameter). But this means also that the exit is most >>> never >> called. >> >> But if you specify the desired time on the job statement (the OP said >> 30 min?), IEFUTL would get called and could take the desired action >> (extend for another 30 after generating some console message?). -- Joel C. Ewing,Bentonville, AR jcew...@acm.org -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
When total CPU time used by the address space is the issue, that can be handled using an IEFUTL exit, setting a default JOB TIME parameter by job class in JES2, and overriding that with TIME parameters on the JOB and EXEC statements as appropriate. The IEFUTL exit can be written to either cancel the job when the limit is exceeded, or ask the operator whether he wants to extend the CPU TIME limit or cancel the job, and differentiate based on Job class, etc. I suspect you could probably write the IEFUTL exit to auto extend the limit for an STC address space, but put out a console message so the Operator would know it was happening, and he could then manually cancel the address space if the behavior and CPU extension was unreasonable. Joel C. Ewing . On 07/04/2018 12:31 AM, saurabh khandelwal wrote: > I would like operator console to be notified when any address space or Job > taking more CPU for longer then any time we specify > > On Wed, Jul 4, 2018 at 8:27 AM, retired mainframer > wrote: >>> -Original Message- >>> From: IBM Mainframe Discussion List On Behalf >>> Of Peter Hunkeler >>> Sent: Tuesday, July 03, 2018 10:10 PM >>> To: IBM-MAIN@LISTSERV.UA.EDU >>> Subject: AW: Re: CPU Utilization >>> >>>> No, IEFUTL is called INSIDE an Address Space as often as specified in >> SMFPRMxx in >>> statement JWT. >>> >>> >>> and it would also be called when the CPU time used exceeds the time >> limit of the job or >>> step (TIME= parameter). But this means also that the exit is most never >> called. >> >> But if you specify the desired time on the job statement (the OP said 30 >> min?), IEFUTL would get called and could take the desired action (extend >> for >> another 30 after generating some console message?). >> >> -- >> 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 > -- Joel C. Ewing,Bentonville, AR jcew...@acm.org -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: AW: Re: CPU Utilization
On 07/04/2018 06:46 AM, Peter Hunkeler wrote: >> But if you specify the desired time on the job statement (the OP said 30 > min?), IEFUTL would get called and could take the desired action (extend for > another 30 after generating some console message?). > > > > > I stand corrected. I had in mind he wanted an alert when some address space > consumes more that a certain amount of CPU *percent*in an interval, not time. > > > -- > Peter Hunkeler > > > > I think the original poster really needs to consider what his objective is. By itself, percentage of CPU used in real time is not a reliable indicator of a problem with a job. A perfectly tuned complex job on a lightly loaded system may consume a high percentage of a physical CPU for an extended time, but this is goodness if it completes in that much less real time and it shouldn't be penalized for being well-tuned. A job in an infinite loop on a heavily-loaded system may not show up with as high of CPU utilization, but can soak up enough remaing CPU resource to raise your 4-hour MSU average and either raise software charges or result in LPAR capping which could then cause performance issues. The least expensive solution is to set "reasonable" (for your installation) default CPU TIME limits on job classes and in cases where the default TIME is inadequate use reasonable TIME limits on JOB and/or EXEC cards to catch job steps that are using much more total CPU time than is expected. You can write your IEFUTL exit to be sensitive to production job classes and allow the Operator to choose to extend the CPU time for production jobs that appear to be doing useful work, unless the total time is completely out of line with historical usage by the job step. Joel C Ewing -- Joel C. Ewing,Bentonville, AR jcew...@acm.org -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
On Wed, 4 Jul 2018 03:24:00 -0500, Vince Getgood wrote: >Saurabh, >How much is your management willing to spend on achieving this requirement? > >From the few posts that I've seen, I'd suggest no-one here does this now, or >has ever done it. If a company wants to monitor it's mainframe resource usage >dynamically, it buys and impliments a product. (BMC Mainview, ASG TMON, IBM's >Omegamon etc). > >You're basically asking us to help you write something that will replace one >of the above - I'd suggest that's not going to happen - at least not for free! > >Also, consider how much resource a home-grown resource monitor would use? >Who monitors the monitor? > Again, if you have RMF at your site, you could something similar to what we have done using RMFM3B to write out to syslog/operlog what our Group Capacity usage are for our various environments. An example: +RMF300I 3B: Processing CPC Report... +MVS1 - RMF301I: 4H Average: 2 +MVS1 - RMF302I: 4H Max: 3 +MVS1 - RMF303I: WLM Capping %: 0.0 +MVS1 - RMF304I: Group Name: GRPDEVT +MVS1 - RMF305I: Group Limit: 9 +MVS1 - RMF306I: Time until Capping: 195 minutes . Using the above info, you could certainly then setup whatever automation package you have to trap on the relevant fields . Have a look in your *.SERBCLS dataset (members ERBM3B and ERBR3CPC) and that will give you a good starting point to set something up to fit your requirements. As Vince said, you are basically asking/wanting someone else to do the leg work for you for free and that ain't going to happen any time soon! Just to provide you with another source of RMF related info/goodies - https://github.com/IBM/IBM-Z-zOS/tree/master/zOS-RMF Roger -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
AW: Re: CPU Utilization
>But if you specify the desired time on the job statement (the OP said 30 min?), IEFUTL would get called and could take the desired action (extend for another 30 after generating some console message?). I stand corrected. I had in mind he wanted an alert when some address space consumes more that a certain amount of CPU *percent*in an interval, not time. -- Peter Hunkeler -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
Saurabh, How much is your management willing to spend on achieving this requirement? From the few posts that I've seen, I'd suggest no-one here does this now, or has ever done it. If a company wants to monitor it's mainframe resource usage dynamically, it buys and impliments a product. (BMC Mainview, ASG TMON, IBM's Omegamon etc). You're basically asking us to help you write something that will replace one of the above - I'd suggest that's not going to happen - at least not for free! Also, consider how much resource a home-grown resource monitor would use? Who monitors the monitor? You need to go back to your management and explain that, with the tools available to you, what they are trying to achieve is not possible without a significant investment in time and money. I'm available (for a fee!) to explain the real world to your management. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
I would like operator console to be notified when any address space or Job taking more CPU for longer then any time we specify On Wed, Jul 4, 2018 at 8:27 AM, retired mainframer wrote: > > -Original Message- > > From: IBM Mainframe Discussion List On Behalf > > Of Peter Hunkeler > > Sent: Tuesday, July 03, 2018 10:10 PM > > To: IBM-MAIN@LISTSERV.UA.EDU > > Subject: AW: Re: CPU Utilization > > > > >No, IEFUTL is called INSIDE an Address Space as often as specified in > SMFPRMxx in > > statement JWT. > > > > > > and it would also be called when the CPU time used exceeds the time > limit of the job or > > step (TIME= parameter). But this means also that the exit is most never > called. > > But if you specify the desired time on the job statement (the OP said 30 > min?), IEFUTL would get called and could take the desired action (extend > for > another 30 after generating some console message?). > > -- > 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: CPU Utilization
> -Original Message- > From: IBM Mainframe Discussion List On Behalf > Of Peter Hunkeler > Sent: Tuesday, July 03, 2018 10:10 PM > To: IBM-MAIN@LISTSERV.UA.EDU > Subject: AW: Re: CPU Utilization > > >No, IEFUTL is called INSIDE an Address Space as often as specified in SMFPRMxx in > statement JWT. > > > and it would also be called when the CPU time used exceeds the time limit of the job or > step (TIME= parameter). But this means also that the exit is most never called. But if you specify the desired time on the job statement (the OP said 30 min?), IEFUTL would get called and could take the desired action (extend for another 30 after generating some console message?). -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
Thanks for reply. We don't have BMC Mainview or IBM Omegamon or any third party product except Netview from ibm. I read about RMFM3B from rmf user guide also but not much help to setup this for CPU usage . Can anybody help me to achieve this task . On Tue, Jul 3, 2018 at 10:22 PM, Feller, Paul wrote: > If you have BMC Mainview or IBM Omegamon (or some other monitoring > software) consider looking at those to monitor and alert you about CPU > usage. Among other things that is part of what they are created for. > > Thanks.. > > Paul Feller > AGT Mainframe Technical Support > > -Original Message- > From: IBM Mainframe Discussion List [mailto:IBM-MAIN@LISTSERV.UA.EDU] On > Behalf Of saurabh khandelwal > Sent: Tuesday, July 03, 2018 6:42 AM > To: IBM-MAIN@LISTSERV.UA.EDU > Subject: CPU Utilization > > Hello Group, > > We have requirement to trigger alert to operator console, once CPU > utilization for any address space or any Job or any running STC reach to > certain pre defined number. Like any job or address space or STC consuming > CPU for more then 30 min etc , operator should get alter on his console. > > > Is there any way to implement this process using rexx, netview or any > other way . > > -- > 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 > -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
AW: Re: CPU Utilization
>No, IEFUTL is called INSIDE an Address Space as often as specified in SMFPRMxx >in statement JWT. and it would also be called when the CPU time used exceeds the time limit of the job or step (TIME= parameter). But this means also that the exit is most never called. And since we're at JWT: I only recently learnt that z/OS V2.1 introduced two new SMFPRMxx parameters called TWT and SWT. So, we can now manage continuous wait time differently for TSO users, STCs and all the rest, i.e. batch jobs, OMVS process address spaces (I'm just guessing that the latter are handled by JWT). -- Peter Hunkeler -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
If you have BMC Mainview or IBM Omegamon (or some other monitoring software) consider looking at those to monitor and alert you about CPU usage. Among other things that is part of what they are created for. Thanks.. Paul Feller AGT Mainframe Technical Support -Original Message- From: IBM Mainframe Discussion List [mailto:IBM-MAIN@LISTSERV.UA.EDU] On Behalf Of saurabh khandelwal Sent: Tuesday, July 03, 2018 6:42 AM To: IBM-MAIN@LISTSERV.UA.EDU Subject: CPU Utilization Hello Group, We have requirement to trigger alert to operator console, once CPU utilization for any address space or any Job or any running STC reach to certain pre defined number. Like any job or address space or STC consuming CPU for more then 30 min etc , operator should get alter on his console. Is there any way to implement this process using rexx, netview or any other way . -- 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: CPU Utilization
This sounds like something that could be done with a product like BMC Mainview for z/OS or ASG TMON for z/OS (or other). I think alerts can be set for CPU usage in these products. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
On Tue, 3 Jul 2018 14:42:21 +0300, saurabh khandelwal wrote: >Hello Group, > >We have requirement to trigger alert to operator console, once CPU >utilization for any address space or any Job or any running STC reach to >certain pre defined number. Like any job or address space or STC consuming >CPU for more then 30 min etc , operator should get alter on his console. > > >Is there any way to implement this process using rexx, netview or any other >way . > Consider REXX EXEC approach, possibly -- sleep/wake-up, invoke SDSF DA command, parse/interrogate results, take action accordingly, iterate. Scott Barry SBBWorks, Inc. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
saurabh khandelwal wrote: >Can IEFUTL exit also provide live CPU usage detail by every job, stc, address >space, omvs etc and send alert to operator console, if any set limit for CPU >is exceeded . No, IEFUTL is called INSIDE an Address Space as often as specified in SMFPRMxx in statement JWT [depending on further selection by exit]. Anyways, you can perhaps code your own IEFUTL to extract the stats from all 'live address space' and their resource usage. I will not recommend that approach. Reason is that you want to avoid excessive overhead when IEFUTL is triggered by a time-out event. Use, as recommended, RMF and its friends. Groete / Greetings Elardus Engelbrecht -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
I read about RMFM3B from below white paper http://www-03.ibm.com/support/techdocs/atsmastr.nsf/c6192fb3a432612485256d970082de57/399938068ea32df786256e2a005e8c8e/$FILE/RMF_WTO_2010.pdf but i am not sure, how to implement this for checking live cpu usage for every stc, jes job , omvs etc and send notification to operator console once threshold value exceed to avoid issue Is there anybody has implemented this facility. On Tue, Jul 3, 2018 at 4:07 PM, Roger Lowe wrote: > On Tue, 3 Jul 2018 14:42:21 +0300, saurabh khandelwal < > venkatkulkarn...@gmail.com> wrote: > > >Hello Group, > > > >We have requirement to trigger alert to operator console, once CPU > >utilization for any address space or any Job or any running STC reach to > >certain pre defined number. Like any job or address space or STC consuming > >CPU for more then 30 min etc , operator should get alter on his console. > > > > > >Is there any way to implement this process using rexx, netview or any > other > >way . > > > If you have RMF installed, you could look at using RMFM3B and zEvent which > could do what you are after > > Roger > > -- > 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: CPU Utilization
Hello, Thanks for reply. Can IEFUTL exit also provide live CPU usage detail by every job, stc, address space, omvs etc and send alert to operator console, if any set limit for CPU is exceeded . On Tue, Jul 3, 2018 at 6:53 PM, retired mainframer wrote: > You might look at IEFUTL in the MVS Installation Exits manual to see if it > can do what you want. > > > -Original Message- > > From: IBM Mainframe Discussion List On Behalf > > Of saurabh khandelwal > > Sent: Tuesday, July 03, 2018 4:42 AM > > To: IBM-MAIN@LISTSERV.UA.EDU > > Subject: CPU Utilization > > > > Hello Group, > > > > We have requirement to trigger alert to operator console, once CPU > > utilization for any address space or any Job or any running STC reach to > > certain pre defined number. Like any job or address space or STC > consuming > > CPU for more then 30 min etc , operator should get alter on his console. > > > > > > Is there any way to implement this process using rexx, netview or any > other > > way . > > -- > 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: CPU Utilization
You might look at IEFUTL in the MVS Installation Exits manual to see if it can do what you want. > -Original Message- > From: IBM Mainframe Discussion List On Behalf > Of saurabh khandelwal > Sent: Tuesday, July 03, 2018 4:42 AM > To: IBM-MAIN@LISTSERV.UA.EDU > Subject: CPU Utilization > > Hello Group, > > We have requirement to trigger alert to operator console, once CPU > utilization for any address space or any Job or any running STC reach to > certain pre defined number. Like any job or address space or STC consuming > CPU for more then 30 min etc , operator should get alter on his console. > > > Is there any way to implement this process using rexx, netview or any other > way . -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
On Tue, 3 Jul 2018 14:42:21 +0300, saurabh khandelwal wrote: >Hello Group, > >We have requirement to trigger alert to operator console, once CPU >utilization for any address space or any Job or any running STC reach to >certain pre defined number. Like any job or address space or STC consuming >CPU for more then 30 min etc , operator should get alter on his console. > > >Is there any way to implement this process using rexx, netview or any other >way . > If you have RMF installed, you could look at using RMFM3B and zEvent which could do what you are after Roger -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
CPU Utilization
Hello Group, We have requirement to trigger alert to operator console, once CPU utilization for any address space or any Job or any running STC reach to certain pre defined number. Like any job or address space or STC consuming CPU for more then 30 min etc , operator should get alter on his console. Is there any way to implement this process using rexx, netview or any other way . -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: large VSAM LSR buffering vs. CPU utilization
> On May 25, 2017, at 3:28 PM, Martin Packerwrote: > > > > Back in the late 1980's there was a "coffee table book" on Data In Memory, > with a bunch of studies in. VSAM LSR was one of them. It was one of the > cases where CPU was SAVED across the range. > > Pure speculation but I doubt that became untrue subsequently. > > Cheers, Martin SNIP— Martin, Just as a follow on and a bit of trivia. When IBM came out with SAMe. We installed it ASAP. We had run many traces and SMF looking too see what would happen when you increased the number of buffers. After the SAMe installation we saw runtimes dramatically reduced by as much as 50 percent. The SRB time rose an imperceptibly. This was due to SAMe changing the default buffers in QSAM from 2 to (7 or 8) its been a long time. We did not see any additional real storage creep either. IF memory serves me correctly. Before SAMe came out our SE had done a study and got a orange book published as a result. My memory says that SAMe costed $52 a month and was well worth it. Yes there were some bugs that resulted but all in all a damn good piece of software. Ed -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: large VSAM LSR buffering vs. CPU utilization
Back in the late 1980's there was a "coffee table book" on Data In Memory, with a bunch of studies in. VSAM LSR was one of them. It was one of the cases where CPU was SAVED across the range. Pure speculation but I doubt that became untrue subsequently. Cheers, Martin Sent from my iPad > On 24 May 2017, at 14:07, Feller, Paul <paul.fel...@transamerica.com> wrote: > > Most of the time I suggest throwing VSAM (and non-VSAM) buffers at jobs when the business users are complaining about run time and don't have time for app tuning. They are happy to accept a little extra CPU for a decrease in run time. > > Thanks.. > > Paul Feller > AGT Mainframe Technical Support > > -Original Message- > From: IBM Mainframe Discussion List [mailto:IBM-MAIN@LISTSERV.UA.EDU] On Behalf Of John McKown > Sent: Wednesday, May 24, 2017 07:59 > To: IBM-MAIN@LISTSERV.UA.EDU > Subject: large VSAM LSR buffering vs. CPU utilization > > A post on another forum has gotten me to wondering about the size & number > of buffers in a VSAM LSR pool versus the CPU utilization needed to manage > them. Around here the idea has always been "the more buffers the better" > and no analysis has ever been done. But in today's I/O environment, I > wondering if this is unconditionally true. Given the advances in CPU and > its outpacing of the speed of even the fastest I/O (not that my shop is in > this situation), I'm wondering if this simplistic rule is still true. I.e. > Is "throw more buffers at an I/O intensive workload" still true. > > -- > Windows. A funny name for a operating system that doesn't let you see > anything. > > Maranatha! <>< > John McKown > > -- > 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-MAINUnless > stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: large VSAM LSR buffering vs. CPU utilization
Most of the time I suggest throwing VSAM (and non-VSAM) buffers at jobs when the business users are complaining about run time and don't have time for app tuning. They are happy to accept a little extra CPU for a decrease in run time. Thanks.. Paul Feller AGT Mainframe Technical Support -Original Message- From: IBM Mainframe Discussion List [mailto:IBM-MAIN@LISTSERV.UA.EDU] On Behalf Of John McKown Sent: Wednesday, May 24, 2017 07:59 To: IBM-MAIN@LISTSERV.UA.EDU Subject: large VSAM LSR buffering vs. CPU utilization A post on another forum has gotten me to wondering about the size & number of buffers in a VSAM LSR pool versus the CPU utilization needed to manage them. Around here the idea has always been "the more buffers the better" and no analysis has ever been done. But in today's I/O environment, I wondering if this is unconditionally true. Given the advances in CPU and its outpacing of the speed of even the fastest I/O (not that my shop is in this situation), I'm wondering if this simplistic rule is still true. I.e. Is "throw more buffers at an I/O intensive workload" still true. -- Windows. A funny name for a operating system that doesn't let you see anything. Maranatha! <>< John McKown -- 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
large VSAM LSR buffering vs. CPU utilization
A post on another forum has gotten me to wondering about the size & number of buffers in a VSAM LSR pool versus the CPU utilization needed to manage them. Around here the idea has always been "the more buffers the better" and no analysis has ever been done. But in today's I/O environment, I wondering if this is unconditionally true. Given the advances in CPU and its outpacing of the speed of even the fastest I/O (not that my shop is in this situation), I'm wondering if this simplistic rule is still true. I.e. Is "throw more buffers at an I/O intensive workload" still true. -- Windows. A funny name for a operating system that doesn't let you see anything. Maranatha! <>< John McKown -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Performance question - handling of max CPU % utilization
Shane Ginnane wrote: Don't. I bloody wish I could do that. Simple as that - I am constantly trying to disavow people of the notion of one number per day. Agreed! One number does not tell the story truthfully. Don't use the LPAR MVS busy numbers. Period. Go to the Partition Data Report and give those numbers to your boss. Nice graph, simple stacked area graph with the CEC capacity as a big black line at the top. As Allan Staller and Barbara Nitz both kindly said, I should look at the CEC and PR/SM data as RMF samples it. I will use LPAR averages, but to show how really busy the CEC is, I will use other measures. But it helps that I handle each CEC separately, I do not combine any data in one big set of data, simply because the different CEC have different type of workload. The RMF Spreadsheet reporter probably does, Indeed, as others said, this is the tool I should look at it. I also experimented with that RMF portal in the weekend which is already proving very useful for my purpose. Many thanks to all for your kind replies! Much appreciated! Groete / Greetings Elardus Engelbrecht -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Performance question - handling of max CPU % utilization
Agree with the stacked graph of LPARs. But one plea: Do it by processor pool for (at least) GCPs and zIIPs. (IFLs might be meaningful, ICFs less likely, zAAPs possibly.) Which takes us away from one number. :-) Cheers, Martin Martin Packer, zChampion, Principal Systems Investigator, Worldwide Banking Center of Excellence, IBM +44-7802-245-584 email: martin_pac...@uk.ibm.com Twitter / Facebook IDs: MartinPacker Blog: https://www.ibm.com/developerworks/mydeveloperworks/blogs/MartinPacker From: Elardus Engelbrecht elardus.engelbre...@sita.co.za To: IBM-MAIN@LISTSERV.UA.EDU Date: 11/05/2015 07:28 Subject:Re: Performance question - handling of max CPU % utilization Sent by:IBM Mainframe Discussion List IBM-MAIN@LISTSERV.UA.EDU Shane Ginnane wrote: Don't. I bloody wish I could do that. Simple as that - I am constantly trying to disavow people of the notion of one number per day. Agreed! One number does not tell the story truthfully. Don't use the LPAR MVS busy numbers. Period. Go to the Partition Data Report and give those numbers to your boss. Nice graph, simple stacked area graph with the CEC capacity as a big black line at the top. As Allan Staller and Barbara Nitz both kindly said, I should look at the CEC and PR/SM data as RMF samples it. I will use LPAR averages, but to show how really busy the CEC is, I will use other measures. But it helps that I handle each CEC separately, I do not combine any data in one big set of data, simply because the different CEC have different type of workload. The RMF Spreadsheet reporter probably does, Indeed, as others said, this is the tool I should look at it. I also experimented with that RMF portal in the weekend which is already proving very useful for my purpose. Many thanks to all for your kind replies! Much appreciated! Groete / Greetings Elardus Engelbrecht -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Performance question - handling of max CPU % utilization
Martin Packer wrote: Agree with the stacked graph of LPARs. But one plea: Do it by processor pool for (at least) GCPs and zIIPs. (IFLs might be meaningful, ICFs less likely, zAAPs possibly.) Agreed! That I already does. Anyways, according to my local IBMer, the zIIP CPU is shared by the DB2 LPARs. No sense trying to mix and match all CPU usage in one stupid magic number. No ICF and zAAPs for now. Which takes us away from one number. :-) It remind me of this little motto: Take only ONE beer or coffee. It is the size of the mug that matters. ;-D Groete / Greetings Elardus Engelbrecht -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Performance question - handling of max CPU % utilization
Hi to all, It is part of my duties to show to management z/OS performance daily and monthly based on SMF RMF records. CPU% utilization averages, memory usage, MSU, I/O rates, transaction rates, etc. are usually easy to present and explain. You get Averages, 90th percentile, absolute maximum/minimum, etc. as measurement criterias. My problem is: I wish to show that our machines are heavily loaded or maximum loaded. My problem is that the different LPARs have their absolute maximum CPU% utilization at different hourly and 30 minute intervals. So LPAR 1 has 100% CPU utilization at 09:00, but LPAR 2 has 95% CPU utilization at say 13:00. If I combine these values for the day, they're sometimes over 100% which is undesirable or difficult to explain. Sometimes I see those max CPU% drifts very far far away from the usual average CPU% utilization on one or more LPARs at a given interval, but not always at the same time. Question: how do you performance guys and gals present those maximums? Or how do you prove that machines are heavily used? Do you use averages of those maximum CPU% utilization or what do you use? Do you combine all the LPARs and then work out the max? Any trending or statistical analysis methods to consider? I'm using RMF and a commercial product (no SAS) to process those SMF RMF records daily and monthly. z/OS v1.13. Many thanks in advance. Groete / Greetings Elardus Engelbrecht -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Performance question - handling of max CPU % utilization
On Thu, 7 May 2015 06:51:42 -0500, Elardus Engelbrecht wrote: If I combine these values for the day, they're sometimes over 100% which is undesirable or difficult to explain. Don't. Simple as that - I am constantly trying to disavow people of the notion of one number per day. RMF is a sampler - you get averaged numbers. (Re-)averaging those over a period like a day renders them effectively meaningless. And that's if you're using the correct metric(s). Don't use the LPAR MVS busy numbers. Period. Go to the Partition Data Report and give those numbers to your boss. Nice graph, simple stacked area graph with the CEC capacity as a big black line at the top. Easy to see (lack of) white space available. If your software doesn't give you this, get rid of it. The RMF Spreadsheet reporter probably does, but it's been too long since I managed to coerce it to behave. Or you can simply export the numbers to LibreOffice Calc and graph it yourself. Shane ... -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Performance question - handling of max CPU % utilization
Question: how do you performance guys and gals present those maximums? Or how do you prove that machines are heavily used? Do you use averages of those maximum CPU% utilization or what do you use? Do you combine all the LPARs and then work out the max? Any trending or statistical analysis methods to consider? I used to use SMF type70 records (PR/SM data) and compiled grafics (SAS, MXG) that show what each lpar used in each interval. Being 'flatlined' at the top of the box certainly showed that the machine was overloaded, despite the actual lpars still showing room. They could not get it because the physical processors were busy elsewhere. I had a separate grafic for ZIIPs, as they had a different 100% - they were much faster than our GCPs. Barbara -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: Performance question - handling of max CPU % utilization
The RMF CPU report should show this information (for each CEC) both at the aggregate level and broken out by LPAR. You might also look at the RMF Spreadsheet Reporter. More info here : http://www-03.ibm.com/systems/z/os/zos/features/rmf/ and http://publibz.boulder.ibm.com/epubs/pdf/erb2ug01.pdf One way to aggregate multiple CECs is to total all processors and normalize to 100% for the manglers HTH, snip It is part of my duties to show to management z/OS performance daily and monthly based on SMF RMF records. CPU% utilization averages, memory usage, MSU, I/O rates, transaction rates, etc. are usually easy to present and explain. You get Averages, 90th percentile, absolute maximum/minimum, etc. as measurement criterias. My problem is: I wish to show that our machines are heavily loaded or maximum loaded. My problem is that the different LPARs have their absolute maximum CPU% utilization at different hourly and 30 minute intervals. So LPAR 1 has 100% CPU utilization at 09:00, but LPAR 2 has 95% CPU utilization at say 13:00. If I combine these values for the day, they're sometimes over 100% which is undesirable or difficult to explain. Sometimes I see those max CPU% drifts very far far away from the usual average CPU% utilization on one or more LPARs at a given interval, but not always at the same time. Question: how do you performance guys and gals present those maximums? Or how do you prove that machines are heavily used? Do you use averages of those maximum CPU% utilization or what do you use? Do you combine all the LPARs and then work out the max? Any trending or statistical analysis methods to consider? I'm using RMF and a commercial product (no SAS) to process those SMF RMF records daily and monthly. z/OS v1.13. /snip -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Watch the wrap! (Was: CPU Utilization)
Allan and anyone quoting an IBM manual URL I was prompted to include this post in IBM-MAIN based on a thread I have been checking in the IBMTCP-L list. I have used a recent example from IBM-MAIN. If you discard all the unnecessary verbiage following - and including - the question mark, ?, you will save your finger-ends from the pointless pounding of the warning. Also, if you are simply going to provide the Table of Contents for the manual, there is no need for the text /CCONTENTS/. http://publibz.boulder.ibm.com/cgi-bin/bookmgr_OS390/BOOKS/ERBZPM90 works just as well as http://publibz.boulder.ibm.com/cgi-bin/bookmgr_OS390/BOOKS/ERBZPM90/CCONTENTS?SHELF=erbzbkb0DN=SC33-7992-10DT=20100712181154 Chris Mason On Wed, 7 Nov 2012 14:41:56 +, Staller, Allan allan.stal...@kbmg.com wrote: snip Also trying to figure out any batch loads that are of not high priority to run over a non-peak hours which might help a bit to it, /snip This is a techno-political question which only the customer can answer snip Any expert advice on the other ways that i can consider to help would be highly appreciated /snip You might want to review the RMF Performance Management Guide found here: http://publibz.boulder.ibm.com/cgi-bin/bookmgr_OS390/BOOKS/ERBZPM90/CCONTENTS?SHELF=erbzbkb0DN=SC33-7992-10DT=20100712181154 (watch the wrap). Google is your friend. If the RMF PM Guide indicates a problem with (for example, RACF), google for RACF Performance (either on the IBM website or WWW). Last but not least, the IBM TECHDOCS website has lots of good info. http://www-03.ibm.com/support/techdocs/atsmastr.nsf/Web/Technotes -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
Workload Manager is your friend. Hmmm - *IF* you understand the (history of the) workload, the SLAs, the expectations of the customer, the politics, ... And WLM itself of course. Take it from me, it is bloody difficult to walk into a site and make sensible determinations that will be accepted. As usual, Barbara has basically covered things. Shane ... -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
Thanks a lot for everyone and appreciate your time and help, I will consider the recommendations given as i move on, Thanks again! On Thu, Nov 8, 2012 at 11:00 AM, nitz-...@gmx.net nitz-...@gmx.net wrote: Not sure how Workload Manager is being used or configured at your shop. From the original description, I am assuming that that installation runs more than one lpar on the box, most probably with lpar weights that reflect importance of the lpars. If the OP happens to work on an lpar that is not a loved one in terms of lpar weight, then almost no amount of WLM tuning will get that lpar service if another lpar is using that service. What's more, an execution velocity goal that is achievable on a loved lpar won't be on an unloved one for the exact same job with the exact same data. If the *box* (all physical cps) is running at 100% cpu (as they are in many smaller shops), then you need to drastically reduce the number of WLM service classes/service class periods, at least in the unloved lpars. Get down to less than 10 service classes, preferably only one per importance. Be prepared to have a fight with just about everyone because they all think they're more important than the rest of the world and they all want to be at least importance1, better in sysstc. Traditionally, everything will cluster in sysstc and importance 1 service classes, with almost nothing in importance 4 and 5. Barbara Nitz -- 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: CPU Utilization
Probably your mainframe capacity enginneer should have this in his Mind.(It really requires a professional approach) Peter. On Thu, Nov 8, 2012 at 3:45 PM, RCG rkcgowda1...@gmail.com wrote: Thanks a lot for everyone and appreciate your time and help, I will consider the recommendations given as i move on, Thanks again! On Thu, Nov 8, 2012 at 11:00 AM, nitz-...@gmx.net nitz-...@gmx.net wrote: Not sure how Workload Manager is being used or configured at your shop. From the original description, I am assuming that that installation runs more than one lpar on the box, most probably with lpar weights that reflect importance of the lpars. If the OP happens to work on an lpar that is not a loved one in terms of lpar weight, then almost no amount of WLM tuning will get that lpar service if another lpar is using that service. What's more, an execution velocity goal that is achievable on a loved lpar won't be on an unloved one for the exact same job with the exact same data. If the *box* (all physical cps) is running at 100% cpu (as they are in many smaller shops), then you need to drastically reduce the number of WLM service classes/service class periods, at least in the unloved lpars. Get down to less than 10 service classes, preferably only one per importance. Be prepared to have a fight with just about everyone because they all think they're more important than the rest of the world and they all want to be at least importance1, better in sysstc. Traditionally, everything will cluster in sysstc and importance 1 service classes, with almost nothing in importance 4 and 5. Barbara Nitz -- 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 -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
Yeah, then sometimes you need outside help to make things better. Perf. Assoc, Watsonwalker.com, even IBM has tuning services. They are for a fee. It might mean spending thousands to save millions in software/hardware upgrades. In a message dated 11/8/2012 3:01:45 A.M. Central Standard Time, ibm-m...@tpg.com.au writes: Take it from me, it is bloody difficult to walk into a site and make sensible determinations that will be accepted. As usual, Barbara has basically covered things. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
Some manglers feel us techies, experienced ones, are a dime a dozen Scott ford www.identityforge.com Tell me and I'll forget; show me and I may remember; involve me and I'll understand. - Chinese Proverb On Nov 7, 2012, at 10:46 PM, Ted MacNEIL eamacn...@yahoo.ca wrote: I don't contribute to this list very much, but I feel I should in this case. Performance/Capacity Management is not learned overnight. Nor is it a subject where a couple of nuggets on a list serve will solve your problems. Is SMP/E learned overnight? Is being a DBA? COBOL programmer? RACF admin? It's a profession and requires a professional! --Original Message-- From: RCG Sender: IBM Mainframe Discussion List To: IBM-MAIN@LISTSERV.UA.EDU ReplyTo: IBM Mainframe Discussion List Subject: CPU Utilization Sent: 7 Nov 2012 09:30 Dear Group, In one of the shop that I work, I have CPU constraint where the utilization is @ 100% most of the times and that is impacting much of the services which is obvious, I am not a Performance / Capacity expert nor thats my day to day work, Though I am still looking for the ways where we can better the utilization and atleast reduce the impact, additional CPU would definitely resolve that, But that is not something the customer is interested with, I am doing some reading to help tune the performance, Also trying to figure out any batch loads that are of not high priority to run over a non-peak hours which might help a bit to it, Any expert advice on the other ways that i can consider to help would be highly appreciated, Thanks much in advance ! Regards, -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN - Ted MacNEIL eamacn...@yahoo.ca Twitter: @TedMacNEIL -- 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: CPU Utilization
snip Also trying to figure out any batch loads that are of not high priority to run over a non-peak hours which might help a bit to it, /snip This is a techno-political question which only the customer can answer snip Any expert advice on the other ways that i can consider to help would be highly appreciated /snip You might want to review the RMF Performance Management Guide found here: http://publibz.boulder.ibm.com/cgi-bin/bookmgr_OS390/BOOKS/ERBZPM90/CCONTENTS?SHELF=erbzbkb0DN=SC33-7992-10DT=20100712181154 (watch the wrap). Google is your friend. If the RMF PM Guide indicates a problem with (for example, RACF), google for RACF Performance (either on the IBM website or WWW). Last but not least, the IBM TECHDOCS website has lots of good info. http://www-03.ibm.com/support/techdocs/atsmastr.nsf/Web/Technotes HTH, -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
Running 100% is not bad. What is bad is missing SLA's. So if you are not missing SLA's, you don't have an immediate problem. If your shop is like most shops, the work load is growing and that can lead to missing SLA's in the future. I would do the following: 1) Look for 'hot spots', first at the line of business level, then at the sub-system level (CICS, DB2, etc), then at the program and transaction level. 2) Make improvements where you can. 3) Do they have any ZIIP engines? If they have any, how are they being utilized? Can you do more with them? Adding GP engines is expensive in two ways. The processor cost AND the increase in software costs. Adding a ZIIP has no effect on the software costs, mostly. 4) go back to step 1 and do it again. When I did performance monitoring at a very large company we just kept doing steps 1, 2 and 4. We didn't have ZIIP's at that time. Find the top ten users of resources and focus in on them. As you make improvements, some will drop off and new ones will rise to the top, which is why step 4 is important. Chris Blaicher Senior Software Engineer, Software Services Syncsort Incorporated 50 Tice Boulevard, Woodcliff Lake, NJ 07677 P: 201-930-8260 | M: 512-627-3803 E: cblaic...@syncsort.com -Original Message- From: IBM Mainframe Discussion List [mailto:IBM-MAIN@LISTSERV.UA.EDU] On Behalf Of RCG Sent: Wednesday, November 07, 2012 8:30 AM To: IBM-MAIN@LISTSERV.UA.EDU Subject: CPU Utilization Dear Group, In one of the shop that I work, I have CPU constraint where the utilization is @ 100% most of the times and that is impacting much of the services which is obvious, I am not a Performance / Capacity expert nor thats my day to day work, Though I am still looking for the ways where we can better the utilization and atleast reduce the impact, additional CPU would definitely resolve that, But that is not something the customer is interested with, I am doing some reading to help tune the performance, Also trying to figure out any batch loads that are of not high priority to run over a non-peak hours which might help a bit to it, Any expert advice on the other ways that i can consider to help would be highly appreciated, Thanks much in advance ! Regards, -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN ATTENTION: - The information contained in this message (including any files transmitted with this message) may contain proprietary, trade secret or other confidential and/or legally privileged information. Any pricing information contained in this message or in any files transmitted with this message is always confidential and cannot be shared with any third parties without prior written approval from Syncsort. This message is intended to be read only by the individual or entity to whom it is addressed or by their designee. If the reader of this message is not the intended recipient, you are on notice that any use, disclosure, copying or distribution of this message, in any form, is strictly prohibited. If you have received this message in error, please immediately notify the sender and/or Syncsort and destroy all copies of this message in your possession, custody or control. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
Appreciate the prompt response and help Chris Allan, I will consider the advises given and look to see those areas of improvement, Thanks again! Regards, RKC On Wed, Nov 7, 2012 at 8:34 PM, Blaicher, Christopher Y. cblaic...@syncsort.com wrote: Running 100% is not bad. What is bad is missing SLA's. So if you are not missing SLA's, you don't have an immediate problem. If your shop is like most shops, the work load is growing and that can lead to missing SLA's in the future. I would do the following: 1) Look for 'hot spots', first at the line of business level, then at the sub-system level (CICS, DB2, etc), then at the program and transaction level. 2) Make improvements where you can. 3) Do they have any ZIIP engines? If they have any, how are they being utilized? Can you do more with them? Adding GP engines is expensive in two ways. The processor cost AND the increase in software costs. Adding a ZIIP has no effect on the software costs, mostly. 4) go back to step 1 and do it again. When I did performance monitoring at a very large company we just kept doing steps 1, 2 and 4. We didn't have ZIIP's at that time. Find the top ten users of resources and focus in on them. As you make improvements, some will drop off and new ones will rise to the top, which is why step 4 is important. Chris Blaicher Senior Software Engineer, Software Services Syncsort Incorporated 50 Tice Boulevard, Woodcliff Lake, NJ 07677 P: 201-930-8260 | M: 512-627-3803 E: cblaic...@syncsort.com -Original Message- From: IBM Mainframe Discussion List [mailto:IBM-MAIN@LISTSERV.UA.EDU] On Behalf Of RCG Sent: Wednesday, November 07, 2012 8:30 AM To: IBM-MAIN@LISTSERV.UA.EDU Subject: CPU Utilization Dear Group, In one of the shop that I work, I have CPU constraint where the utilization is @ 100% most of the times and that is impacting much of the services which is obvious, I am not a Performance / Capacity expert nor thats my day to day work, Though I am still looking for the ways where we can better the utilization and atleast reduce the impact, additional CPU would definitely resolve that, But that is not something the customer is interested with, I am doing some reading to help tune the performance, Also trying to figure out any batch loads that are of not high priority to run over a non-peak hours which might help a bit to it, Any expert advice on the other ways that i can consider to help would be highly appreciated, Thanks much in advance ! Regards, -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN ATTENTION: - The information contained in this message (including any files transmitted with this message) may contain proprietary, trade secret or other confidential and/or legally privileged information. Any pricing information contained in this message or in any files transmitted with this message is always confidential and cannot be shared with any third parties without prior written approval from Syncsort. This message is intended to be read only by the individual or entity to whom it is addressed or by their designee. If the reader of this message is not the intended recipient, you are on notice that any use, disclosure, copying or distribution of this message, in any form, is strictly prohibited. If you have received this message in error, please immediately notify the sender and/or Syncsort and destroy all copies of this message in your possession, custody or control. -- 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: CPU Utilization
Don't take this the wrong way, but performance analysis and tuning isn't something you pick up in your spare time. You need to get a professional involved. That's not me, so this isn't self-interested. On Wed, Nov 7, 2012 at 10:04 AM, Blaicher, Christopher Y. cblaic...@syncsort.com wrote: Running 100% is not bad. What is bad is missing SLA's. So if you are not missing SLA's, you don't have an immediate problem. If your shop is like most shops, the work load is growing and that can lead to missing SLA's in the future. I would do the following: 1) Look for 'hot spots', first at the line of business level, then at the sub-system level (CICS, DB2, etc), then at the program and transaction level. 2) Make improvements where you can. 3) Do they have any ZIIP engines? If they have any, how are they being utilized? Can you do more with them? Adding GP engines is expensive in two ways. The processor cost AND the increase in software costs. Adding a ZIIP has no effect on the software costs, mostly. 4) go back to step 1 and do it again. When I did performance monitoring at a very large company we just kept doing steps 1, 2 and 4. We didn't have ZIIP's at that time. Find the top ten users of resources and focus in on them. As you make improvements, some will drop off and new ones will rise to the top, which is why step 4 is important. Chris Blaicher Senior Software Engineer, Software Services Syncsort Incorporated 50 Tice Boulevard, Woodcliff Lake, NJ 07677 P: 201-930-8260 | M: 512-627-3803 E: cblaic...@syncsort.com -Original Message- From: IBM Mainframe Discussion List [mailto:IBM-MAIN@LISTSERV.UA.EDU] On Behalf Of RCG Sent: Wednesday, November 07, 2012 8:30 AM To: IBM-MAIN@LISTSERV.UA.EDU Subject: CPU Utilization Dear Group, In one of the shop that I work, I have CPU constraint where the utilization is @ 100% most of the times and that is impacting much of the services which is obvious, I am not a Performance / Capacity expert nor thats my day to day work, Though I am still looking for the ways where we can better the utilization and atleast reduce the impact, additional CPU would definitely resolve that, But that is not something the customer is interested with, I am doing some reading to help tune the performance, Also trying to figure out any batch loads that are of not high priority to run over a non-peak hours which might help a bit to it, Any expert advice on the other ways that i can consider to help would be highly appreciated, Thanks much in advance ! Regards, -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN ATTENTION: - The information contained in this message (including any files transmitted with this message) may contain proprietary, trade secret or other confidential and/or legally privileged information. Any pricing information contained in this message or in any files transmitted with this message is always confidential and cannot be shared with any third parties without prior written approval from Syncsort. This message is intended to be read only by the individual or entity to whom it is addressed or by their designee. If the reader of this message is not the intended recipient, you are on notice that any use, disclosure, copying or distribution of this message, in any form, is strictly prohibited. If you have received this message in error, please immediately notify the sender and/or Syncsort and destroy all copies of this message in your possession, custody or control. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN -- zMan -- I've got a mainframe and I'm not afraid to use it -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
Workload Manager is your friend. Cram on this before attempting anything else. It is your primary tool to accomplish your objectives, but you need to understand how the tool works. If you don't know which end of the hammer to hold you'll never drive in the nail. If you don't have SAS and MXG, then get them. You will need something to research, measure and validate your results. SAS for Windows workstation is relatively cheap, and can easily handle tens of GB of SMF data. Don't believe anyone that says you can do this efficiently with the RMF spreadsheet reporter and a few CBT programs. Ron -Original Message- From: IBM Mainframe Discussion List [mailto:IBM-MAIN@LISTSERV.UA.EDU] On Behalf Of RCG Sent: Wednesday, November 07, 2012 6:30 AM To: IBM-MAIN@LISTSERV.UA.EDU Subject: [IBM-MAIN] CPU Utilization Dear Group, In one of the shop that I work, I have CPU constraint where the utilization is @ 100% most of the times and that is impacting much of the services which is obvious, I am not a Performance / Capacity expert nor thats my day to day work, Though I am still looking for the ways where we can better the utilization and atleast reduce the impact, additional CPU would definitely resolve that, But that is not something the customer is interested with, I am doing some reading to help tune the performance, Also trying to figure out any batch loads that are of not high priority to run over a non-peak hours which might help a bit to it, Any expert advice on the other ways that i can consider to help would be highly appreciated, Thanks much in advance ! Regards, -- 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: CPU Utilization
Is the mainframe used for production or development? If it's used for development then SimpList could help you free up some CPU cycles. Developers spend much of their time moving from one ISPF panel or vendor product to another, and every command they enter to travel from one place to another causes a mainframe interrupt. They might also use 3.4 to access data sets, which means the mainframe is constantly being asked to perform catalog searches. Even simple things like changing the attributes of a data set requires a number of mainframe interrupts. With all of these things, having the CPU at 100% can cause developers to spend long periods of time staring at the clock. SimpList helps resolve this problem by providing a fully integrated workbench where many tasks can be performed with just one or two mouse clicks, thereby sending a minimum number of interrupts to the mainframe. This is a very inexpensive way to reduce CPU cycles as well as saving developers huge amounts of time. Feel free to contact me off-line if you have any questions. Dave Salt SimpList(tm) - try it, you'll get it! http://www.mackinney.com/products/program-development/simplist.html Date: Wed, 7 Nov 2012 10:04:08 -0500 From: cblaic...@syncsort.com Subject: Re: CPU Utilization To: IBM-MAIN@LISTSERV.UA.EDU Running 100% is not bad. What is bad is missing SLA's. So if you are not missing SLA's, you don't have an immediate problem. If your shop is like most shops, the work load is growing and that can lead to missing SLA's in the future. I would do the following: 1) Look for 'hot spots', first at the line of business level, then at the sub-system level (CICS, DB2, etc), then at the program and transaction level. 2) Make improvements where you can. 3) Do they have any ZIIP engines? If they have any, how are they being utilized? Can you do more with them? Adding GP engines is expensive in two ways. The processor cost AND the increase in software costs. Adding a ZIIP has no effect on the software costs, mostly. 4) go back to step 1 and do it again. When I did performance monitoring at a very large company we just kept doing steps 1, 2 and 4. We didn't have ZIIP's at that time. Find the top ten users of resources and focus in on them. As you make improvements, some will drop off and new ones will rise to the top, which is why step 4 is important. Chris Blaicher Senior Software Engineer, Software Services Syncsort Incorporated 50 Tice Boulevard, Woodcliff Lake, NJ 07677 P: 201-930-8260 | M: 512-627-3803 E: cblaic...@syncsort.com -Original Message- From: IBM Mainframe Discussion List [mailto:IBM-MAIN@LISTSERV.UA.EDU] On Behalf Of RCG Sent: Wednesday, November 07, 2012 8:30 AM To: IBM-MAIN@LISTSERV.UA.EDU Subject: CPU Utilization Dear Group, In one of the shop that I work, I have CPU constraint where the utilization is @ 100% most of the times and that is impacting much of the services which is obvious, I am not a Performance / Capacity expert nor thats my day to day work, Though I am still looking for the ways where we can better the utilization and atleast reduce the impact, additional CPU would definitely resolve that, But that is not something the customer is interested with, I am doing some reading to help tune the performance, Also trying to figure out any batch loads that are of not high priority to run over a non-peak hours which might help a bit to it, Any expert advice on the other ways that i can consider to help would be highly appreciated, Thanks much in advance ! Regards, -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN ATTENTION: - The information contained in this message (including any files transmitted with this message) may contain proprietary, trade secret or other confidential and/or legally privileged information. Any pricing information contained in this message or in any files transmitted with this message is always confidential and cannot be shared with any third parties without prior written approval from Syncsort. This message is intended to be read only by the individual or entity to whom it is addressed or by their designee. If the reader of this message is not the intended recipient, you are on notice that any use, disclosure, copying or distribution of this message, in any form, is strictly prohibited. If you have received this message in error, please immediately notify the sender and/or Syncsort and destroy all copies of this message in your possession, custody or control. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
As has been stated before, Capacity and Performance is not a casual thing that you can do over lunch. That being said, you can look at a few things. 100% is not a bad thing, modern mainframes are designed to run at 100%. The question is whether things are being delayed. If you have access to RMF data, you can look at the PI (performance index) of individual service classes over a certain interval. This is much easier if you have a tool other than RMF, but you have to use what you have. If the PIs are greater than 1.0, you're being delayed and that's where you're problem is residing. Not sure how Workload Manager is being used or configured at your shop. Check the coefficients value. Oddly enough, a fair number of shops take the default values (which are wrong) instead of the recommended (which should be the default, but they're not). Not sure what type of box you're using (i.e. z9, z10, etc). Is there a defined capacity set, and is that capacity set higher or lower than the initialization values? This can wreck all sorts of havoc when the system gets busy. Are your catalogs loaded into memory? Kind of difficult to determine where your problem might lie, because it literally could be anything. Adding processors may or may not help, and could actually make things worse. Depending on your box type, increasing the number of processors without increasing the number of MSUs (and thus changing your classification) will lower the number of MIPS each processor will execute per second. If you're running DB2 or think a zIIP might help, you can set PROJECTCPU to YES (in IEAOPT00) and find out. You can also check the values for the percentage of I/O interruptions to be processed through the test pending interrupt. It should be 10,30. Also ERV - then number of CPU service units and address space can absorb when it's possibly causing enqueue contentions. The default is 500, which might be too low for your shop. Good luck. Anne R. Adams DTI, Systems Engineering (302) 298 - 3196 -Original Message- From: IBM Mainframe Discussion List [mailto:IBM-MAIN@LISTSERV.UA.EDU] On Behalf Of RCG Sent: Wednesday, November 07, 2012 9:30 AM To: IBM-MAIN@LISTSERV.UA.EDU Subject: CPU Utilization Dear Group, In one of the shop that I work, I have CPU constraint where the utilization is @ 100% most of the times and that is impacting much of the services which is obvious, I am not a Performance / Capacity expert nor thats my day to day work, Though I am still looking for the ways where we can better the utilization and atleast reduce the impact, additional CPU would definitely resolve that, But that is not something the customer is interested with, I am doing some reading to help tune the performance, Also trying to figure out any batch loads that are of not high priority to run over a non-peak hours which might help a bit to it, Any expert advice on the other ways that i can consider to help would be highly appreciated, Thanks much in advance ! Regards, -- 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: CPU Utilization
I don't contribute to this list very much, but I feel I should in this case. Performance/Capacity Management is not learned overnight. Nor is it a subject where a couple of nuggets on a list serve will solve your problems. Is SMP/E learned overnight? Is being a DBA? COBOL programmer? RACF admin? It's a profession and requires a professional! --Original Message-- From: RCG Sender: IBM Mainframe Discussion List To: IBM-MAIN@LISTSERV.UA.EDU ReplyTo: IBM Mainframe Discussion List Subject: CPU Utilization Sent: 7 Nov 2012 09:30 Dear Group, In one of the shop that I work, I have CPU constraint where the utilization is @ 100% most of the times and that is impacting much of the services which is obvious, I am not a Performance / Capacity expert nor thats my day to day work, Though I am still looking for the ways where we can better the utilization and atleast reduce the impact, additional CPU would definitely resolve that, But that is not something the customer is interested with, I am doing some reading to help tune the performance, Also trying to figure out any batch loads that are of not high priority to run over a non-peak hours which might help a bit to it, Any expert advice on the other ways that i can consider to help would be highly appreciated, Thanks much in advance ! Regards, -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN - Ted MacNEIL eamacn...@yahoo.ca Twitter: @TedMacNEIL -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: CPU Utilization
Not sure how Workload Manager is being used or configured at your shop. From the original description, I am assuming that that installation runs more than one lpar on the box, most probably with lpar weights that reflect importance of the lpars. If the OP happens to work on an lpar that is not a loved one in terms of lpar weight, then almost no amount of WLM tuning will get that lpar service if another lpar is using that service. What's more, an execution velocity goal that is achievable on a loved lpar won't be on an unloved one for the exact same job with the exact same data. If the *box* (all physical cps) is running at 100% cpu (as they are in many smaller shops), then you need to drastically reduce the number of WLM service classes/service class periods, at least in the unloved lpars. Get down to less than 10 service classes, preferably only one per importance. Be prepared to have a fight with just about everyone because they all think they're more important than the rest of the world and they all want to be at least importance1, better in sysstc. Traditionally, everything will cluster in sysstc and importance 1 service classes, with almost nothing in importance 4 and 5. Barbara Nitz -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN