I think it's also time to discuss LPAR vs basic, and why basic is no longer provided. LPAR is a good thing. I am running 7 lpars and the lpar sie overhead, while greater than 0%, is tough to distinguish. This is due to what I view as two factors: 1. the speed of the z9-109 instruction processing, aka MIPs. Don't be jealous if you don't have one! 2. the microcode on the z9-109 is excellent. Not that I have seen it but it is fast and slick. Also the additional i/o processors, microcode assists on a z9 are great. I would have to see the manuals with the striped covers and then my eyes will buron out of my head, but the LPAR SIE is a souped up SIE - I guess. Also the dispatching alogrithms match z/VM but must be written in microcode or close to it. It ain't the 3330-x imbedded disk drive with vm370 ala the 1980's! Maybe it's the cp scheduler and dispatcher in microcode, maybe it isn't, maybe it's close, maybe it's an approximation. Whatever it's darn fast.
So losing basic mode is no loss as far as I am concerned. I think the z9-109 is an over achiever. My I/O rates scream to IBM and HDS DASD devices (unless, um, they start to flashcopy every volume durin PRIME SHIFT - come on boys). You can get into trouble with LPARs and shoot yourself in the foot with settings, etc. but there is no way of telling if this is what happened to Jim at the DR site. It's the 3rd level dispatches. You can measure Java etc. and i think it's worthwhile, but the linux cycle under z/vm under z/vm is pain. An intercept from SIE means the virtual machine did something in it's slice that can't be EMULATED (processor speed) so it has to be SIMULATED (many instructions to accomplish one virtual machine instruction). Losing V=R, well, that may be something to pine for, but most, a lot, of z/os shops opt for LPAR instead of preferred guests. Most but not all. I have not said there is no need for V=R (Ask your friendly VSE shop). Endicott can only get so much included, and running linux is a high priority. David Bill Bitner wrote:
I do not believe %SIC means what you think it means. I had a conversation about this in the past I think with someone from the County if it is the same configuration. At the DR location, if you are running Linux as a guest of VM and VM running as a guest of another VM system and all that running in an LPAR, then you effectively have 3 layers of SIE. (1 for each VM and 1 for LPAR). Current machines can handle 2 layers of SIE very efficiently, but performance does drop off significantly with 3 layers of SIE. SIE is how VM (and LPAR) dispatches work. We sometimes use the phrase SIE Break or Exit from SIE to describe the cases where the virtual (or logical) processor stops running under SIE and control is passed back to the hipervisor. There are two reasons to exit SIE: intercepts and interrupts. The %SIC is what percentage of the exits were for intercepts. So the number could be high here because of exits for SIEs done under the three levels, but it could also be influenced by CPU bound work or work that naturally has low level of interrupts. Bill Bitner - VM Performance Evaluation - IBM Endicott - 607-429-3286 ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
