We have a separation policy here. Development and test cannot touch the production network and vice versa. The separation is by LPAR in that there is one production LPAR running on a machine that is otherwise development. Other than that, the separation is by cpu serial number. We do not separate by data center, though. Our big, important apps have their own playgrounds and are separated by miles of land and even oceans from each other. The development box shares a data center with 2 big apps plus some number of z/OS systems.
Regards, Richard Schuh -----Original Message----- From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Marcy Cortes Sent: Monday, May 21, 2007 3:22 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance Rules of Thumb ? Some very good points Alan! Some workload is just fine at 100%. One thing to note about 100% is how far over it you are :) (and your perf monitor might help you figure that out!) Something is needing more when you are at 100 - whether it puts up with getting less than what it needs is certainly app dependent. You can control some of that with priorities and then maybe set share absolute limithard for the one that goes in a tight loop... If you are on top of that... or maybe you could automate that ... We have a big important app that is very cpu intensive but also CPU sensitive. If it isn't able to process its tran within a certain time amount, that tran is stood in for by another system and screens go red, the world is paged, and long running whose-fault conference calls start. We certainly don't want joe users's leaky java program testing to cause that to happen. So, those guys are on different z/VM lpars (actually, different data centers, but that's a whole 'nother isolate test from production story - more z/OS's fuss than ours, but we play within the rules :). The 60% is because we have 3 lpars, they want the work load of 1 to fit on the other 2 if one should go down (yes, VM systems do go down and HW could go down too :). Someone here wants 6 lpars, then we can run them at 75-80% and be able to have whitespace for 1. Not sure that extreme is a good idea either, at least for the VM sysprogs who have enough to do as it is. I'd be curious to know if others also use extreme separation or if they find they can get test/dev to play well with prod. ?? Marcy Cortes "This message may contain confidential and/or privileged information. If you are not the addressee or authorized to receive this for the addressee, you must not use, copy, disclose, or take any action based on this message or any information herein. If you have received this message in error, please advise the sender immediately by reply e-mail and delete this message. Thank you for your cooperation." -----Original Message----- From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Alan Ackerman Sent: Monday, May 21, 2007 2:42 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: [IBMVM] Performance Rules of Thumb ? On Mon, 21 May 2007 11:26:20 -0700, Lionel B. Dyck <[EMAIL PROTECTED]>= wrote: >Thanks - you're points on YMMV is very appropo - I was just looking for >= >starting points. > >Lionel B. Dyck, Consultant/Specialist >Enterprise Platform Services, Mainframe Engineering KP-IT Enterprise >Engineering, Client and Platform Engineering Services = >(CAPES) >925-926-5332 (8-473-5332) | E-Mail: [EMAIL PROTECTED] >AIM: lbdyck | Yahoo IM: lbdyck >Kaiser Service Credo: "Our cause is health. Our passion is service. >We?r= e >here to make lives better.? I think rules of thumb are a bad starting point. They either give you false sense of security, or you waste time trying to track down non- existent problems. We have run VM systems at 100% CPU utilization without problems -- but it= was because we had enough low-priority work to sop up the cycles. With = a "lumpier" workload (such as Linux guests), you might get in trouble somewhere between 80-100%. The 60% that someone cited seems extreme, but = if you need to hold cycles in reserve, then you do. I/O and paging rates sustainable depend largely on the capability of the = I/O subsystem. But long before you approach that capacity, you will start= seeing problems due to queuing on a device. Adding more paging device addresses may help, for example. Some queuing is hidden though -- it exists but you will never see it. Clearly, you cannot run at above 100% of the CPU capacity or the I/O capacity. But it is very workload dependent on how close you can get to = 100%.