Re: GRS RESMIL setting on CPU consumption
On Wed, 23 Apr 2014 04:56:28 -0500, Anthony Fletcher flet...@nz1.ibm.com wrote: It is a while since there had been any discussion of appropriate values of RESMIL and its effect on CPU consumptiom. It appears that there was general concensus that RESMIL(OFF) would be a bad idea, and that RESMIL(0) would be best since it should be managed dynamically. A blast from (my) past: https://listserv.ua.edu/cgi-bin/wa?A2=IBM-MAIN;8d626696.0802 Scott Fagen Chief Architect - System z CA Technologies -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: GRS RESMIL setting on CPU consumption
Hello, I asked Charlie Favell in z/OS BCP System Test about RESMIL... A slightly different perspective. Are you using MIM's MII to manage global ENQs and by association GRSRNL=EXCLUDE? If so, global ENQs are a rare thing and you shouldn't be concerned about performance. Maybe try RESMIL(10). If not, you may be okay with just RESMIL(5). GRS will tune it down automatically when the RSA is heavily loaded, but most of the time it is sent empty! JG Joe Gentile z/OS GRS and Logger Development (845)435-2184 (T/L 295-2184) jwgen...@us.ibm.com -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: GRS RESMIL setting on CPU consumption
Hello, Based on the lack of warnings in the following page in the GRS Planning Guide ( http://pic.dhe.ibm.com/infocenter/zos/v2r1/index.jsp?topic=%2Fcom.ibm.zos.v2r1.ieag400%2Ftunring.htm ), and some customer experience, I don't think that RESMIL=0 or OFF will severely impact your CPU utilization. I'll run that by a colleague. Always best to try it in a test environment first of course. Ring mode performance for Global ENQs is still drastically slower in Ring mode than it is in Star. The following performance tools for ENQs (also documented in GRS Planning) could be helpful in addition to RMF Monitor III reports: SYS1.SAMPLIB(ISGNQRSP) - Measures Global ENQ performance GRS EQDQ Monitor, ISGAUDIT - Provides a more comprehensive look at ENQ data including performance, with some performance overhead Please check out the references attached to this email for more information about GRS performance. GRS Planning Guide: Ring info http://pic.dhe.ibm.com/infocenter/zos/v2r1/topic/com.ibm.zos.v2r1.ieag400/gring.htm http://pic.dhe.ibm.com/infocenter/zos/v2r1/topic/com.ibm.zos.v2r1.ieag400/tunring.htm GRS EQDQ Monitor (ISGAUDIT) http://pic.dhe.ibm.com/infocenter/zos/v2r1/topic/com.ibm.zos.v2r1.ieag400/emon.htm GRS ISGNQRSP Info http://pic.dhe.ibm.com/infocenter/zos/v2r1/topic/com.ibm.zos.v2r1.ieag400/visype.htm#visype Introduction to GRS SHARE Presentation from 2010 http://www.share.org/p/do/sd/topic=60sid=591 (need to register at share.org) IBM GRS presentation at NASPAA (which can be found at http://www.google.com/url?sa=trct=jq=esrc=ssource=webcd=3ved=0CDwQFjACurl=ftp%3A%2F%2Fpublic.dhe.ibm.com%2Fs390%2Fzos%2Fracf%2Fpdf%2Fnynaspa_2005_11_29_grs_101.pdfei=bhoGU5DcKKfH0QHYq4GIDAusg=AFQjCNEBpznPzbj7oHh_sOUyyIFXDhr5dQsig2=lU02rfp3PM6Xj5f58V0hMQbvm=bv.61725948,d.dmQcad=rja ) z/OS GRS: Performance Considerations White Paper (which can be found at http://www-03.ibm.com/support/techdocs/atsmastr.nsf/5cb5ed706d254a8186256c71006d2e0a/e6be7f9c4c2704578625777000715888/$FILE/grs_perf_v11.pdf ) Thanks, JG Joe Gentile z/OS GRS and Logger Development (845)435-2184 (T/L 295-2184) jwgen...@us.ibm.com -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: GRS RESMIL setting on CPU consumption
Correction: There are some cautionary statements about RESMIL=OFF regarding CPU consumption. Zero is different as GRS may still do tuning. JG Joe Gentile z/OS GRS and Logger Development (845)435-2184 (T/L 295-2184) jwgen...@us.ibm.com - Forwarded by Joseph W Gentile/Poughkeepsie/IBM on 04/27/2014 09:14 PM - From: Joseph W Gentile/Poughkeepsie/IBM To: IBM-MAIN@LISTSERV.UA.EDU, Date: 04/27/2014 09:09 PM Subject:Re: GRS RESMIL setting on CPU consumption Hello, Based on the lack of warnings in the following page in the GRS Planning Guide ( http://pic.dhe.ibm.com/infocenter/zos/v2r1/index.jsp?topic=%2Fcom.ibm.zos.v2r1.ieag400%2Ftunring.htm ), and some customer experience, I don't think that RESMIL=0 or OFF will severely impact your CPU utilization. I'll run that by a colleague. Always best to try it in a test environment first of course. Ring mode performance for Global ENQs is still drastically slower in Ring mode than it is in Star. The following performance tools for ENQs (also documented in GRS Planning) could be helpful in addition to RMF Monitor III reports: SYS1.SAMPLIB(ISGNQRSP) - Measures Global ENQ performance GRS EQDQ Monitor, ISGAUDIT - Provides a more comprehensive look at ENQ data including performance, with some performance overhead Please check out the references attached to this email for more information about GRS performance. GRS Planning Guide: Ring info http://pic.dhe.ibm.com/infocenter/zos/v2r1/topic/com.ibm.zos.v2r1.ieag400/gring.htm http://pic.dhe.ibm.com/infocenter/zos/v2r1/topic/com.ibm.zos.v2r1.ieag400/tunring.htm GRS EQDQ Monitor (ISGAUDIT) http://pic.dhe.ibm.com/infocenter/zos/v2r1/topic/com.ibm.zos.v2r1.ieag400/emon.htm GRS ISGNQRSP Info http://pic.dhe.ibm.com/infocenter/zos/v2r1/topic/com.ibm.zos.v2r1.ieag400/visype.htm#visype Introduction to GRS SHARE Presentation from 2010 http://www.share.org/p/do/sd/topic=60sid=591 (need to register at share.org) IBM GRS presentation at NASPAA (which can be found at http://www.google.com/url?sa=trct=jq=esrc=ssource=webcd=3ved=0CDwQFjACurl=ftp%3A%2F%2Fpublic.dhe.ibm.com%2Fs390%2Fzos%2Fracf%2Fpdf%2Fnynaspa_2005_11_29_grs_101.pdfei=bhoGU5DcKKfH0QHYq4GIDAusg=AFQjCNEBpznPzbj7oHh_sOUyyIFXDhr5dQsig2=lU02rfp3PM6Xj5f58V0hMQbvm=bv.61725948,d.dmQcad=rja ) z/OS GRS: Performance Considerations White Paper (which can be found at http://www-03.ibm.com/support/techdocs/atsmastr.nsf/5cb5ed706d254a8186256c71006d2e0a/e6be7f9c4c2704578625777000715888/$FILE/grs_perf_v11.pdf ) Thanks, JG Joe Gentile z/OS GRS and Logger Development (845)435-2184 (T/L 295-2184) jwgen...@us.ibm.com -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: GRS RESMIL setting on CPU consumption
Anthony, If you have DB2 in that plex, I would think twice (or thrice) about increasing the resmil value, as DB2 does heavy enqueue activity. I remember one presentation (I think it was at share or maybe another conference), where someone from IBM GRS presented the following item: Company A implemented GRS Star and opened a ticket with ibm as a db2 application, that normally ran 9 hours, ran in just a few minutes, so they were sure something had gone wrong. Working with ibm, they went through testing, first back on GRS RING and the default of RESMIL=10. Result: the application ran 9 hours IBM had them change to RESMIL(1) (or maybe 0) and the elapsed time dropped to 90 minutes Change to GRS Star, and the application ran in 9 minutes. Just an example of the impact of the RESMIL setting (and even of what GRS Star can give (yes it does require the Coupling Facility). Peter -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: GRS RESMIL setting on CPU consumption
Yes RESMIL(5) will reduce cpu consumption, it will also potentially cause performance delays to your jobs, as each system will hold the enq request for a minimum of 5 ms. If you have any applications that generate alot of enqueues, that need to be sent through the ring, they will run longer. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
Re: GRS RESMIL setting on CPU consumption
On Wed, 23 Apr 2014 04:56:28 -0500, Anthony Fletcher wrote: It is a while since there had been any discussion of appropriate values of RESMIL and its effect on CPU consumptiom. It appears that there was general concensus that RESMIL(OFF) would be a bad idea, and that RESMIL(0) would be best since it should be managed dynamically. In a 3-LPAR plex with GRS linking the 3 (not via VF) then surely whatever happens in one LPAR will happen in the other two, but does that mean that if GRS dynamically sets the RESMIL to 0 it will cause increased CPU consumption in all 3 LPARS? Might it be better to set RESMIL to, say 5 so that it never goes too far down? H. From a long association of messing with GRS ring, I've come to the conclusion that it doesn't pay to screw with GRS ring. Given your employer, you may have better access to any (apparently non-existent) lower level tools to see what is actually happening in a ring, but otherwise, it may be best to leave well enough alone. Changes to ring parameters can be difficult to remediate in need - although 3-system is certainly better than some I've seen. If you are seeing increased CPU consumption in GRS address space(s), treat that as a symptom of some other (major ?) problem, not the base problem. Rings can be very sensitive to things like small partitions - sandpit getting single digit CPU share say due to weights. And maybe even just one CPU to make it worse. Empirical evidence suggests capping events can severely adversely impact all the systems in a ring. I had protracted discussions with GRS support about this, but they were unable/unwilling to provide tools that would allow evidence to be collected. Maybe you will have better luck. Shane ... -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
GRS RESMIL setting on CPU consumption
It is a while since there had been any discussion of appropriate values of RESMIL and its effect on CPU consumptiom. It appears that there was general concensus that RESMIL(OFF) would be a bad idea, and that RESMIL(0) would be best since it should be managed dynamically. In a 3-LPAR plex with GRS linking the 3 (not via VF) then surely whatever happens in one LPAR will happen in the other two, but does that mean that if GRS dynamically sets the RESMIL to 0 it will cause increased CPU consumption in all 3 LPARS? Might it be better to set RESMIL to, say 5 so that it never goes too far down? -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN