Re: Hyper PAVs vs. Dynamic PAVs
Yes. The I/O that needs an Alias will get one with HyperPAV. With Dynamic it may have to wait (depending on I/O concurrency to that UCB and current number of PAVs assigned). Additional benifts. A reduction in the number of Alias addresses that you need to define per LCU (Allows for more Base addresses!) No WLM intra-Sysplex communication needed (each LPAR needs to know the current assignment status). Alias assignment is based on I/O priority John - Original Message - From: gsg [EMAIL PROTECTED] Newsgroups: bit.listserv.ibm-main Sent: Friday, September 12, 2008 10:11 PM Subject: Re: Hyper PAVs vs. Dynamic PAVs Is there significant improvement from Dynamic to Hyper? -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hyper PAVs vs. Dynamic PAVs
I have a paper that documents the reaction time of WLM based PAV as opposed to HyperPAV. HyperPAV is instantaneous, in fact the IOSQ time in my test was zero for all intervals. Dynamic PAV controlled through WLM takes four minutes to adjust to a changing workload for a device and never gets the IOSQ to zero. The paper has been accepted for presentation at CMG in December of this year. Tom Moulder -Original Message- From: IBM Mainframe Discussion List [mailto:[EMAIL PROTECTED] On Behalf Of Mark Zelden Sent: Friday, September 12, 2008 3:19 PM To: IBM-MAIN@BAMA.UA.EDU Subject: Re: Hyper PAVs vs. Dynamic PAVs On Fri, 12 Sep 2008 19:57:01 +, Ted MacNEIL [EMAIL PROTECTED] wrote: One of the nice things about HyperPAV is that if you share DASD between sysplexes, you don't have to worry about PAV thrashing like you would if you had WLM controlled PAVs active to both sysplexes for the same DASD. IBM has always admitted, unashamedly, that as of SYSPLEX, DASD sharing is problematic between SYSPLEX'. Also, for the exaggerator in the crowd, the maximum WLM interval is 10s, not 10m. - 10 seconds is the policy adjustment interval. But there are 2 algorithms controlling PAVs IIRC. One is to help goals and that runs about every 30 seconds. The other is for overall system efficiency and that runs about every 60 seconds. The problem is that they don't do much at each interval (each one is slightly different) so adjustments are very slow. Multiple devices can be adjusted, but no more than one alias will be added to a particular device during that cycle. So it can take a long time (maybe even 10 minutes) to get to where you need if there is a sustained increase in IOSQ. A good example would be when your onlines come up in the morning or when everyone walks into the office around the same time and starts doing work. I probably have this close... but not exact (it's been a while since I reviewed this information). It is documented in the Sysprog's Guide to WLM RedBook. Mark -- Mark Zelden Sr. Software and Systems Architect - z/OS Team Lead Zurich North America / Farmers Insurance Group - ZFUS G-ITO mailto:[EMAIL PROTECTED] z/OS Systems Programming expert at http://expertanswercenter.techtarget.com/ Mark's MVS Utilities: http://home.flash.net/~mzelden/mvsutil.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Hyper PAVs vs. Dynamic PAVs
Can anyone tell me what kind of improvements can be realized changing PAVs from Dynamic to Hyper? What is the real difference between the two? TIA -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hyper PAVs vs. Dynamic PAVs
In a message dated 9/12/2008 2:03:47 P.M. Central Daylight Time, [EMAIL PROTECTED] writes: Can anyone tell me what kind of improvements can be realized changing PAVs from Dynamic to Hyper? I'll leave the quantification to others and to your own mileage calculations. What is the real difference between the two? First came static PAVs. You decide where the PAVs are inside each LCU (Logical Control Unit), a collection of up to 256 devices. If you want to change any of them, you have to redo your configuration. Lots of work. Error-prone PITA. Hard to change. You must know your hot devices in advance. Next came dynamic PAV. The WLM decides within one LCU what should be PAVed for the next WLM interval in order to deliver your requested goals, WLM issues control I/O commands to the controller, PAV array is reset inside the controller, and the new PAV configuration is fixed in concrete (static) until the end of the next WLM interval, maybe 10 minutes later? Next came HyperPAV. IOS decides on an I/O by I/O basis if a PAV is needed for the next I/O, finds one from a pool of available PAV UCBs, directs the new I/O to a PAVed UCB which the controller knows how to send to the proper device, then IOS returns the PAV UCB to the pool of available PAV UCBs when the I/O ends. You don't have to reconfig. You don't wait until the end of the WLM interval. IOS does no control I/O to tell the controller a new PAV configuration. Instantaneously dynamic as opposed to quasi-static as opposed to seriously static. Bill Fairchild Rocket Software **Pt...Have you heard the news? There's a new fashion blog, plus the latest fall trends and hair styles at StyleList.com. (http://www.stylelist.com/trends?ncid=aolsty000514) -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hyper PAVs vs. Dynamic PAVs
On Fri, 12 Sep 2008 15:24:15 EDT, IBM Mainframe Discussion List [EMAIL PROTECTED] wrote: Next came HyperPAV. IOS decides on an I/O by I/O basis if a PAV is needed for the next I/O, finds one from a pool of available PAV UCBs, directs the new I/O to a PAVed UCB which the controller knows how to send to the proper device, then IOS returns the PAV UCB to the pool of available PAV UCBs when the I/O ends. You don't have to reconfig. You don't wait until the end of the WLM interval. IOS does no control I/O to tell the controller a new PAV configuration. Instantaneously dynamic as opposed to quasi-static as opposed to seriously static. One of the nice things about HyperPAV is that if you share DASD between sysplexes, you don't have to worry about PAV thrashing like you would if you had WLM controlled PAVs active to both sysplexes for the same DASD. Mark -- Mark Zelden Sr. Software and Systems Architect - z/OS Team Lead Zurich North America / Farmers Insurance Group - ZFUS G-ITO mailto:[EMAIL PROTECTED] z/OS Systems Programming expert at http://expertanswercenter.techtarget.com/ Mark's MVS Utilities: http://home.flash.net/~mzelden/mvsutil.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hyper PAVs vs. Dynamic PAVs
One of the nice things about HyperPAV is that if you share DASD between sysplexes, you don't have to worry about PAV thrashing like you would if you had WLM controlled PAVs active to both sysplexes for the same DASD. IBM has always admitted, unashamedly, that as of SYSPLEX, DASD sharing is problematic between SYSPLEX'. Also, for the exaggerator in the crowd, the maximum WLM interval is 10s, not 10m. - Too busy driving to stop for gas! -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hyper PAVs vs. Dynamic PAVs
In a message dated 9/12/2008 2:57:34 P.M. Central Daylight Time, [EMAIL PROTECTED] writes: Also, for the exaggerator in the crowd, the maximum WLM interval is 10s, not 10m. I wasn't exaggerating. I had no idea of its magnitude, and was guessing some number of minutes because of RMF's interval unit of granularity. HyperPAV cuts the interval down to the millisecond level, which is how long it takes for one I/O to run to completion. WLM's maximum of 10 seconds is worth about 10,000 such I/O requests. Bill Fairchild Rocket Software **Pt...Have you heard the news? There's a new fashion blog, plus the latest fall trends and hair styles at StyleList.com. (http://www.stylelist.com/trends?ncid=aolsty000514) -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hyper PAVs vs. Dynamic PAVs
Is there significant improvement from Dynamic to Hyper? -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hyper PAVs vs. Dynamic PAVs
On Fri, 12 Sep 2008 19:57:01 +, Ted MacNEIL [EMAIL PROTECTED] wrote: One of the nice things about HyperPAV is that if you share DASD between sysplexes, you don't have to worry about PAV thrashing like you would if you had WLM controlled PAVs active to both sysplexes for the same DASD. IBM has always admitted, unashamedly, that as of SYSPLEX, DASD sharing is problematic between SYSPLEX'. Also, for the exaggerator in the crowd, the maximum WLM interval is 10s, not 10m. - 10 seconds is the policy adjustment interval. But there are 2 algorithms controlling PAVs IIRC. One is to help goals and that runs about every 30 seconds. The other is for overall system efficiency and that runs about every 60 seconds. The problem is that they don't do much at each interval (each one is slightly different) so adjustments are very slow. Multiple devices can be adjusted, but no more than one alias will be added to a particular device during that cycle. So it can take a long time (maybe even 10 minutes) to get to where you need if there is a sustained increase in IOSQ. A good example would be when your onlines come up in the morning or when everyone walks into the office around the same time and starts doing work. I probably have this close... but not exact (it's been a while since I reviewed this information). It is documented in the Sysprog's Guide to WLM RedBook. Mark -- Mark Zelden Sr. Software and Systems Architect - z/OS Team Lead Zurich North America / Farmers Insurance Group - ZFUS G-ITO mailto:[EMAIL PROTECTED] z/OS Systems Programming expert at http://expertanswercenter.techtarget.com/ Mark's MVS Utilities: http://home.flash.net/~mzelden/mvsutil.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
