Re: Dedicated vs. Shared CPs
There WAS a corner case and at least one customer I knew of had it: With ICFs you could expand into shared engines - with the non expansion case being all dedicated. That went away some time ago. I think what you said about z10 corroborates my memory that Dynamic ICF Expansion was only supported on z9 and prior. (Fair mucked up my CPU reporting and as it was going anyway I deigned not to fix my code.) :-) Cheers, Martin Martin Packer, Mainframe Performance Consultant, zChampion 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: Skip Robinson jo.skip.robin...@sce.com To: IBM-MAIN@bama.ua.edu, Date: 16/02/2012 05:07 Subject: Re: Dedicated vs. Shared CPs Sent by: IBM Mainframe Discussion List IBM-MAIN@bama.ua.edu I checked out an MVS Image profile on my (brand spanking new!) z196. It looks the choice is between dedicated or shared CPs. A CF LPAR offers more choices, but oddly fewer choices than the z10 it replaced. . . JO.Skip Robinson SCE Infrastructure Technology Services Electric Dragon Team Paddler SHARE MVS Program Co-Manager 626-302-7535 Office 323-715-0595 Mobile jo.skip.robin...@sce.com From: Barbara Nitz nitz-...@gmx.net To: IBM-MAIN@bama.ua.edu Date: 02/15/2012 08:58 PM Subject:Re: Dedicated vs. Shared CPs Sent by:IBM Mainframe Discussion List IBM-MAIN@bama.ua.edu Is it possible to mix shared and dedicated CPs on the same LPAR? No. Not on a z9 and not the way you mean. The RMF report deals with what an lpar can look like when it uses Hiperdispatch. That is not available on a z9. Hiperdispatch semi-dedicates logical processors to physical processors depending on the workload. And most probably depending on the number of physical cps. (Which is why we asked for the number of your physical cps in the other thread - you haven't answered that!) Once hiperdispatch is on, a physical cp gets a weight, which can be low, medium or high. High amounts to being (semi-)dedicated. Does anyone run mixed like this, a dedicated number to cover the minimum expected MSU of an LPAR, then some logicals to float between LPARS? Everyone who uses hiperdipatch. Does anyone know what is needed in terms of outages/lpar resets to move from a completely logical CP environment to a mixed environment? An upgrade to a machine supporting hiperdispatch. Which means at least one IPL. But as you can see from my questions starting this thread, it is possible that you cannot use hiperdispatch - whenever your logical to physical cp ratio is so bad that 4 or more logical cps compete for one physical, due to the number of lpars you have, for instance. In the very first presentation I heard (by Bob Rogers) about hiperdispatch, he said that it will turn itself off if the ratio is really bad. I haven't heard that confirmed anywhere, though. And coming from a z9 presumably to a z196, chances are very good that you would loose physical cps to keep money down. Try making your bosses understand that they cannot use the same number of logicals when the number of physicals decreases Barbara Nitz -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.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...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
Thanks Barbera, that explains 'mixed' perfectly. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
Is it possible to mix shared and dedicated CPs on the same LPAR? I thought it wasn't, but a note on the RMF partition data report implies it is; On WGT Either the partition's current weighting of the shared processor resources or one of the following indicators: DED Indicates that the partition is dedicated. DMX Indicates that a mix of dedicated and non-dedicated processors is used in this partition. WMX Indicates that different share values are assigned to logical processors used in this partition Does anyone run mixed like this, a dedicated number to cover the minimum expected MSU of an LPAR, then some logicals to float between LPARS? Does anyone know what is needed in terms of outages/lpar resets to move from a completely logical CP environment to a mixed environment? Thanks, Joe -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
Is it possible to mix shared and dedicated CPs on the same LPAR? No. Not on a z9 and not the way you mean. The RMF report deals with what an lpar can look like when it uses Hiperdispatch. That is not available on a z9. Hiperdispatch semi-dedicates logical processors to physical processors depending on the workload. And most probably depending on the number of physical cps. (Which is why we asked for the number of your physical cps in the other thread - you haven't answered that!) Once hiperdispatch is on, a physical cp gets a weight, which can be low, medium or high. High amounts to being (semi-)dedicated. Does anyone run mixed like this, a dedicated number to cover the minimum expected MSU of an LPAR, then some logicals to float between LPARS? Everyone who uses hiperdipatch. Does anyone know what is needed in terms of outages/lpar resets to move from a completely logical CP environment to a mixed environment? An upgrade to a machine supporting hiperdispatch. Which means at least one IPL. But as you can see from my questions starting this thread, it is possible that you cannot use hiperdispatch - whenever your logical to physical cp ratio is so bad that 4 or more logical cps compete for one physical, due to the number of lpars you have, for instance. In the very first presentation I heard (by Bob Rogers) about hiperdispatch, he said that it will turn itself off if the ratio is really bad. I haven't heard that confirmed anywhere, though. And coming from a z9 presumably to a z196, chances are very good that you would loose physical cps to keep money down. Try making your bosses understand that they cannot use the same number of logicals when the number of physicals decreases Barbara Nitz -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
I checked out an MVS Image profile on my (brand spanking new!) z196. It looks the choice is between dedicated or shared CPs. A CF LPAR offers more choices, but oddly fewer choices than the z10 it replaced. . . JO.Skip Robinson SCE Infrastructure Technology Services Electric Dragon Team Paddler SHARE MVS Program Co-Manager 626-302-7535 Office 323-715-0595 Mobile jo.skip.robin...@sce.com From: Barbara Nitz nitz-...@gmx.net To: IBM-MAIN@bama.ua.edu Date: 02/15/2012 08:58 PM Subject:Re: Dedicated vs. Shared CPs Sent by:IBM Mainframe Discussion List IBM-MAIN@bama.ua.edu Is it possible to mix shared and dedicated CPs on the same LPAR? No. Not on a z9 and not the way you mean. The RMF report deals with what an lpar can look like when it uses Hiperdispatch. That is not available on a z9. Hiperdispatch semi-dedicates logical processors to physical processors depending on the workload. And most probably depending on the number of physical cps. (Which is why we asked for the number of your physical cps in the other thread - you haven't answered that!) Once hiperdispatch is on, a physical cp gets a weight, which can be low, medium or high. High amounts to being (semi-)dedicated. Does anyone run mixed like this, a dedicated number to cover the minimum expected MSU of an LPAR, then some logicals to float between LPARS? Everyone who uses hiperdipatch. Does anyone know what is needed in terms of outages/lpar resets to move from a completely logical CP environment to a mixed environment? An upgrade to a machine supporting hiperdispatch. Which means at least one IPL. But as you can see from my questions starting this thread, it is possible that you cannot use hiperdispatch - whenever your logical to physical cp ratio is so bad that 4 or more logical cps compete for one physical, due to the number of lpars you have, for instance. In the very first presentation I heard (by Bob Rogers) about hiperdispatch, he said that it will turn itself off if the ratio is really bad. I haven't heard that confirmed anywhere, though. And coming from a z9 presumably to a z196, chances are very good that you would loose physical cps to keep money down. Try making your bosses understand that they cannot use the same number of logicals when the number of physicals decreases Barbara Nitz -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
Barbara Nitz writes: And coming from a z9 presumably to a z196, chances are very good that you would loose physical cps to keep money down. It's not clear yet which z9 the original poster has. Moving from a z9 BC to a z114, no. There are more capacity models in the z114 and more configurable engines available but with the same capacity starting points, so there's more choice, not less. Moreover, a z114 could be an appropriate upgrade from a smaller z9 EC. It's very possible a z114 could provide *more* engines with the same overall PCI. Moving from a z9 EC to a z196, maybe, but there's still more overall flexibility. The z9 EC only permits up to the first eight engines as sub-capacity engines, while the z196 supports up to 15 sub-capacity engines. There's also a double MSU technology dividend in that move. (Yes, there's a technology dividend from z10 EC to z196, too -- a bit in MSU terms plus AWLC.) And presumably the original poster would be moving to AWLC on the z196, meaning sub-capacity licensing would be available, if he doesn't already have that. Money is overwhelmingly correlated with peak utilization, not capacity -- and with a very curvaceous curve past the base investment. If the original poster has a 6xx or 7xx z9 EC, then it's possible that they could move to a z196 with more engines in a 4xx or perhaps 5xx configuration. For example, if they've got a z9 EC 703 (1607 PCI, 229 MSUs), they could move to a z114 Y03 (1788 PCI), W04 (1595), X04 (1941), V05 (1723). They could also move to a z196 408 (1667) or 503 (1642). Every one of those options except the X04 would have fewer MSUs than the z9 EC 703. And every one of those options would have AWLC or AEWLC in addition to the MSU change. Note that both the z114 and z196 support HiperDispatch. The best thing to do is to sit down with a system architect or specialist to determine the right fit for the workloads. But, to net it out, reducing the number of engines is an unlikely *requirement*, even for money reasons. Timothy Sipples Resident Enterprise Architect (Based in Singapore) E-Mail: timothy.sipp...@us.ibm.com -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Logical-to-Physical CP ratios, was: Re: Dedicated vs. Shared CPs
Last I heard from IBM and other sources, the official recommendation was not to exceed 2:1 logical to physical. I myself, have run with 3:1 logical to physical with similar results to yours on 9672, z/800, z/9 and z/10(no hiperdispatch). This seems to be a constant! No info on z/196 w/hiperdispatch or higher ratios... snip. This is a handy question. How is your logical-to-physical CP ratio? And how high is your 'lpar overhead' (type70 SMF) ? We run on z196 and have a 18 logicals to 4 physicals ratio (4.5:1), with lpar overhead on these (slowed down) GCPs reaching 2% at the most. The other box runs 12:4 (3:1), with lpar overhead around 0.8%. No Hiperdispatch. Would Hiperdispatch even run in a 4,5:1 ratio environment? Or would it throw in the towel and turn itself off? We also run IFLs, and the picture here is drastically different. On the IFLs we run 18 logicals on 8 physicals and 20:8 (2,5:1) on the other box. The IFLs are obviously not slowed down. LPAR overhead for IFLs on the 20:8 box is near 3% at times, and around 1.5% on the other. Has anyone else any numbers for a z196 with similar logical-to-physical ratios? I am always surprised about how high lpar overhead on the IFLs is. Is VM doing anything extra with regard to lpar overhead that an MVS wouldn't do? (The linuxes running under each VM are also grossly overspecified in their 'logical' processor specs, which *should* show up as overhead within that VM lpar (vm monitor). I have been wondering if some of that overhead is given out to general lpar overhead. Or is it just that the high speed of the IFLs (compared to our slowed-down GCPs) causes this overhead? Thanks, Barbara NItz /snip -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
Bob, Would you know of any redbooks/manuals that might discuss this? -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
I do not. Sorry. Check the past SHARE Proceedings, particularly for Kath Walsh's sessions. Although I know of nothing specific, she has talked about topics such as this in the past. Bob Shannon Rocket Software -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Dedicated vs. Shared CPs
I know you can define processors as dedicated or shared. Is there any advantages/disadvantages to dedicating some to specific LPARs and shared for others? TIA -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
Hi GSG, Simply, if you dedicate some CP to a busy lpar, it can improve performance - if that is what the lpar needs. If you dedicate all that a lpar is likely to use at peak untilitzation, cp reaources are wasted when the lpar is not so busy. Many shops use a mix of shared and dedicated cp based on the needs and priorities of their specific workloads and hardware configurations . Before changing cp assignments, be sure that is what needs to be done. HTH, Linda - Original Message - From: gsg gsg_...@yahoo.com To: IBM-MAIN@bama.ua.edu Sent: Wednesday, February 8, 2012 11:25:48 AM Subject: Dedicated vs. Shared CPs I know you can define processors as dedicated or shared. Is there any advantages/disadvantages to dedicating some to specific LPARs and shared for others? TIA -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
Using HiperDispatch on a z10 or z196 provides many of the benefits of dedicated processors with the flexibility of shared processors. Thanks, Sam -Original Message- From: IBM Mainframe Discussion List [mailto:IBM-MAIN@bama.ua.edu] On Behalf Of gsg Sent: Wednesday, February 08, 2012 2:26 PM To: IBM-MAIN@bama.ua.edu Subject: Dedicated vs. Shared CPs I know you can define processors as dedicated or shared. Is there any advantages/disadvantages to dedicating some to specific LPARs and shared for others? TIA This email/fax message is for the sole use of the intended recipient(s) and may contain confidential and privileged information. Any unauthorized review, use, disclosure or distribution of this email/fax is prohibited. If you are not the intended recipient, please destroy all paper and electronic copies of the original message. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
Our workload is entirely for development. Since we can't predict where demand will come from we share all of the CPs so that capacity will float to where it's needed. One disadvantage is that this is bad for the cache(s). In our environment the cost is acceptable. I wouldn't do this if we ran a production workload. Bob Shannon Rocket Software -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
Bob, Why would this be bad for cache? -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Re: Dedicated vs. Shared CPs
Why would this be bad for cache? Since multiple LPAR are isong the CP, because multiple LPARs are causing cache updates. If you have production LPARs it's best to limit cache updates to a single LPAR so that your cache hit ratio will be better. Dedicating CPs will do that. Bob Shannon Rocket Software -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
Logical-to-Physical CP ratios, was: Re: Dedicated vs. Shared CPs
Our workload is entirely for development. Since we can't predict where demand will come from we share all of the CPs so that capacity will float to where it's needed. One disadvantage is that this is bad for the cache(s). In our environment the cost is acceptable. I wouldn't do this if we ran a production workload. This is a handy question. How is your logical-to-physical CP ratio? And how high is your 'lpar overhead' (type70 SMF) ? We run on z196 and have a 18 logicals to 4 physicals ratio (4.5:1), with lpar overhead on these (slowed down) GCPs reaching 2% at the most. The other box runs 12:4 (3:1), with lpar overhead around 0.8%. No Hiperdispatch. Would Hiperdispatch even run in a 4,5:1 ratio environment? Or would it throw in the towel and turn itself off? We also run IFLs, and the picture here is drastically different. On the IFLs we run 18 logicals on 8 physicals and 20:8 (2,5:1) on the other box. The IFLs are obviously not slowed down. LPAR overhead for IFLs on the 20:8 box is near 3% at times, and around 1.5% on the other. Has anyone else any numbers for a z196 with similar logical-to-physical ratios? I am always surprised about how high lpar overhead on the IFLs is. Is VM doing anything extra with regard to lpar overhead that an MVS wouldn't do? (The linuxes running under each VM are also grossly overspecified in their 'logical' processor specs, which *should* show up as overhead within that VM lpar (vm monitor). I have been wondering if some of that overhead is given out to general lpar overhead. Or is it just that the high speed of the IFLs (compared to our slowed-down GCPs) causes this overhead? Thanks, Barbara NItz -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
