Re: Page Dataset Question
IIRC, Isn't it suggested to allocate one Local on a full volume? On Thu, Jan 21, 2010 at 2:04 AM, Staller, Allan allan.stal...@kbm1.comwrote: This is a performance issue. In order for ASM not to care, the pages are distributed (roughly equally) across all available page datsets. If the algorithm were to take into account the available space: 1) the additional overhead of keep track (probably miniscue per IO, but imagine a paging rate of 100's or even thousands per second. (this actually happened in the good old days). 2) The concentration of pages in the larger page dataset would create a hot spot in the aux stor subsystem and provide uneven performance, depending on where your stolen pages were. snip Wouldn't it make more sense if distributed equally were defined as a percentage of available space rather than number of pages? /snip All this does is allow for human error if the page ds's are not sized equally! -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html -- Cobe Xu Best Regards --- zOS Performance Capacity Analyst E2E Performance Analyst Email: cob...@gmail.com --- -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Page Dataset Question
Cobe, I've never seen that suggested or recommended. As Mr Merrill says Just because you can, doesn't mean you should. Ron -Original Message- From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of Cobe Xu Sent: Thursday, January 21, 2010 8:17 AM To: IBM-MAIN@bama.ua.edu Subject: Re: [IBM-MAIN] Page Dataset Question IIRC, Isn't it suggested to allocate one Local on a full volume? On Thu, Jan 21, 2010 at 2:04 AM, Staller, Allan allan.stal...@kbm1.comwrote: This is a performance issue. In order for ASM not to care, the pages are distributed (roughly equally) across all available page datsets. If the algorithm were to take into account the available space: 1) the additional overhead of keep track (probably miniscue per IO, but imagine a paging rate of 100's or even thousands per second. (this actually happened in the good old days). 2) The concentration of pages in the larger page dataset would create a hot spot in the aux stor subsystem and provide uneven performance, depending on where your stolen pages were. snip Wouldn't it make more sense if distributed equally were defined as a percentage of available space rather than number of pages? /snip All this does is allow for human error if the page ds's are not sized equally! -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html -- Cobe Xu Best Regards --- zOS Performance Capacity Analyst E2E Performance Analyst Email: cob...@gmail.com --- -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Page Dataset Question
Since the ASM change to support large page data sets (z/OS 1.10, rolled back to 1.8 1.9 via APAR OA20749), we've been using full volume 3390-9 for locals and in some of our very large LPARs full volume 3390-27 volumes. Since we couldn't support WLM PAVs in some environments (shared DASD between sysplexes), we've never had more than one local per volume. And as someone already mentioned, HIPERPAV doesn't address this (unless that code was fixed in 1.11 - I don't recall). I think was we migrated DASD the ROT was to put it on 3390-27 as long as there were at least 4 of them (to be the equivalent of the former space allocated on 3390-3). Mark -- Mark Zelden Sr. Software and Systems Architect - z/OS Team Lead Zurich North America / Farmers Insurance Group - ZFUS G-ITO mailto:mark.zel...@zurichna.com z/OS Systems Programming expert at http://expertanswercenter.techtarget.com/ Mark's MVS Utilities: http://home.flash.net/~mzelden/mvsutil.html On Thu, 21 Jan 2010 09:07:16 -0800, Ron Hawkins ron.hawkins1...@sbcglobal.net wrote: Cobe, I've never seen that suggested or recommended. As Mr Merrill says Just because you can, doesn't mean you should. Ron -Original Message- From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of Cobe Xu Sent: Thursday, January 21, 2010 8:17 AM To: IBM-MAIN@bama.ua.edu Subject: Re: [IBM-MAIN] Page Dataset Question IIRC, Isn't it suggested to allocate one Local on a full volume? On Thu, Jan 21, 2010 at 2:04 AM, Staller, Allan allan.stal...@kbm1.comwrote: This is a performance issue. In order for ASM not to care, the pages are distributed (roughly equally) across all available page datsets. If the algorithm were to take into account the available space: 1) the additional overhead of keep track (probably miniscue per IO, but imagine a paging rate of 100's or even thousands per second. (this actually happened in the good old days). 2) The concentration of pages in the larger page dataset would create a hot spot in the aux stor subsystem and provide uneven performance, depending on where your stolen pages were. snip Wouldn't it make more sense if distributed equally were defined as a percentage of available space rather than number of pages? /snip All this does is allow for human error if the page ds's are not sized equally! -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html -- Cobe Xu Best Regards --- zOS Performance Capacity Analyst E2E Performance Analyst Email: cob...@gmail.com --- -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu 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 lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Page Dataset Question
snip We had a problem with a system with not enough page space defined. I remember discussions a long time ago on IBM-Main saying that if you have local page datasets of differing sizes, that you will never use all of the space on the larger page datasets if you add a larger page dataset. Is that still true? We have 3 locals and the PLPA and Common all allocated on the same 3390 Mod 3 pack. We just added a 4th page dataset that is a whole 3390 volume. Obviously, slightly larger than all the other locals. We actually had to IPL the system, as one of the locals went over 80% full, and the system wouldn't allow address space create. Since we IPL'd with the same 3 page datasets, one of the locals is already at 59% full. We page added the 4th dataset after the system was up, and it is still at 0%. /snip Generally, ASM will utilize the page datasets equally, so if they differ in size, the smaller one's will have a higher percent utilization. It sounds like you only had one local at the time. When it reached the 80% threshold, address space creation was stopped. The smallest local PAGEDS is the control. The fact that the 4th dataset is at 0% indicated the pages on aux storage are from long term tasks (e.g. dfHSM). Where the task is up for days/months at a time. Usually, this has been the SMF30 records being retained for eventual write to SMF. It also indicates that there is not much pressure on real storage. All of that being said, my recommendation would be to have at least 2 locals of nearly equal size. The first local would be specified in IEASYSxx and the second would be included via a PAGEADD in COMMNDxx. This will occur early enough in the IPL that both locals will be used equally. The local PAGEds's should be sized so that the max slot utilization does not exceed 30% on any individual dataset. This defeat the block paging algorithm and reduces the efficiency of ASM. HTH, -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Page Dataset Question
Thanks Allan. That pretty much confirms what I thought, that the paging was distributed equally among all of the page datasets. We had 3 local page datasets. One of them is significantly smaller than the other 2, which is the one that went over 80%. This is a tech support sandbox system with no real work on it. Thanks for the info. Eric -- Eric Bielefeld Systems Programmer IBM MVS Technical Services Dubuque, Iowa 563-845-4363 Staller wrote: Generally, ASM will utilize the page datasets equally, so if they differ in size, the smaller one's will have a higher percent utilization. It sounds like you only had one local at the time. When it reached the 80% threshold, address space creation was stopped. The smallest local PAGEDS is the control. The fact that the 4th dataset is at 0% indicated the pages on aux storage are from long term tasks (e.g. dfHSM). Where the task is up for days/months at a time. Usually, this has been the SMF30 records being retained for eventual write to SMF. It also indicates that there is not much pressure on real storage. All of that being said, my recommendation would be to have at least 2 locals of nearly equal size. The first local would be specified in IEASYSxx and the second would be included via a PAGEADD in COMMNDxx. This will occur early enough in the IPL that both locals will be used equally. The local PAGEds's should be sized so that the max slot utilization does not exceed 30% on any individual dataset. This defeat the block paging algorithm and reduces the efficiency of ASM. HTH, -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Page Dataset Question
Eric, Since it is a sandbox, why not do a pagedelete against the small one and see what happens to the other 3 datasets? You should see some activity going onto the new large one. Rex -Original Message- From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of Eric Bielefeld Sent: Wednesday, January 20, 2010 9:56 AM To: IBM-MAIN@bama.ua.edu Subject: Re: Page Dataset Question Thanks Allan. That pretty much confirms what I thought, that the paging was distributed equally among all of the page datasets. We had 3 local page datasets. One of them is significantly smaller than the other 2, which is the one that went over 80%. This is a tech support sandbox system with no real work on it. Thanks for the info. Eric -- Eric Bielefeld Systems Programmer IBM MVS Technical Services Dubuque, Iowa 563-845-4363 Staller wrote: Generally, ASM will utilize the page datasets equally, so if they differ in size, the smaller one's will have a higher percent utilization. It sounds like you only had one local at the time. When it reached the 80% threshold, address space creation was stopped. The smallest local PAGEDS is the control. The fact that the 4th dataset is at 0% indicated the pages on aux storage are from long term tasks (e.g. dfHSM). Where the task is up for days/months at a time. Usually, this has been the SMF30 records being retained for eventual write to SMF. It also indicates that there is not much pressure on real storage. All of that being said, my recommendation would be to have at least 2 locals of nearly equal size. The first local would be specified in IEASYSxx and the second would be included via a PAGEADD in COMMNDxx. This will occur early enough in the IPL that both locals will be used equally. The local PAGEds's should be sized so that the max slot utilization does not exceed 30% on any individual dataset. This defeat the block paging algorithm and reduces the efficiency of ASM. HTH, -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Page Dataset Question
On Wed, 20 Jan 2010 15:56:15 +, Eric Bielefeld wrote: Thanks Allan. That pretty much confirms what I thought, that the paging was distributed equally among all of the page datasets. We had 3 local page datasets. One of them is significantly smaller than the other 2, which is the one that went over 80%. This is a tech support sandbox system with no real work on it. Thanks for the info. Wouldn't it make more sense if distributed equally were defined as a percentage of available space rather than number of pages? (Yes, I know that for disparate data set sizes this would distribute the I/Os less uniformly, but it would delay the ultimate disaster.) -- gil -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Page Dataset Question
It definitely would from a utilization standpoint, but my guess is that there would be higher overhead in the paging algorithms to keep track of this. And we all want paging to happen as quickly as possible so my guess is that IBM decided to go with simplicity and speed. Granted in Eric's case where there are 4-5 page datasets on a single volume, I'm not sure speed is a valid issue... Rex -Original Message- From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of Paul Gilmartin Sent: Wednesday, January 20, 2010 11:08 AM To: IBM-MAIN@bama.ua.edu Subject: Re: Page Dataset Question On Wed, 20 Jan 2010 15:56:15 +, Eric Bielefeld wrote: Thanks Allan. That pretty much confirms what I thought, that the paging was distributed equally among all of the page datasets. We had 3 local page datasets. One of them is significantly smaller than the other 2, which is the one that went over 80%. This is a tech support sandbox system with no real work on it. Thanks for the info. Wouldn't it make more sense if distributed equally were defined as a percentage of available space rather than number of pages? (Yes, I know that for disparate data set sizes this would distribute the I/Os less uniformly, but it would delay the ultimate disaster.) -- gil -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Page Dataset Question
This is a performance issue. In order for ASM not to care, the pages are distributed (roughly equally) across all available page datsets. If the algorithm were to take into account the available space: 1) the additional overhead of keep track (probably miniscue per IO, but imagine a paging rate of 100's or even thousands per second. (this actually happened in the good old days). 2) The concentration of pages in the larger page dataset would create a hot spot in the aux stor subsystem and provide uneven performance, depending on where your stolen pages were. snip Wouldn't it make more sense if distributed equally were defined as a percentage of available space rather than number of pages? /snip All this does is allow for human error if the page ds's are not sized equally! -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Page Dataset Question
We had a problem with a system with not enough page space defined. I remember discussions a long time ago on IBM-Main saying that if you have local page datasets of differing sizes, that you will never use all of the space on the larger page datasets if you add a larger page dataset. Is that still true? We have 3 locals and the PLPA and Common all allocated on the same 3390 Mod 3 pack. We just added a 4th page dataset that is a whole 3390 volume. Obviously, slightly larger than all the other locals. We actually had to IPL the system, as one of the locals went over 80% full, and the system wouldn't allow address space create. Since we IPL'd with the same 3 page datasets, one of the locals is already at 59% full. We page added the 4th dataset after the system was up, and it is still at 0%. -- Eric Bielefeld Systems Programmer IBM MVS Technical Services Dubuque, Iowa 563-845-4363 -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html