Crispin, >From what I read of your email you are trying to choose between two sub-capacity z10 models, one with two CP and one with four, that have roughly the same total MIPS as your current z9 which has two CP. If I have that part correct, then the sub-capacity uniprocessor MIPS speed of the z10 with two CP will be approximately the same as your current z9, and the sub-capacity uniprocessor MIPS speed of the z10 with four CP will half of the current z9.
There have been several emails that paraphrase what I said about your 4xCP z10: "any compute bound task that runs at close to 100% of one CP is going to take twice as long" but I'll explain what I meant. Any batch job, SQL Query, transaction, etc that is compute bound usually means that the speed of the CPU governs the elapsed time of the unit of work. Where the UOW has little in the way of Multi-tasking, like a batch compile or a single DB2 thread, that governor will be the uniprocessor speed, and not the total capacity of your z9 or z10. If go from 2xCP to 4xCP that are 50% slower, your compute bound tasks will take twice as long to complete because they are running at half the MIPS speed. In the case of your development environment, if you are currently running enough concurrent work, or UOWs with unusually high multitasking, and your 2xCP z9 running at close to 100% busy most of the time, then one of three things will happen if you go to 4xCP that are 50% slower: a) Net throughput stays the same as a high number of tasks compete for a small number of CP and net capacity has stayed the same. b) Net throughput reduces because the multitasking level occasionally (or often) falls below the number of CP and the slower uniprocessor retards the CPU execution speed. A symptom of this is the inability to no longer sustain 100% busy for the same workload mix. c) The throughput of some UOWs improves, while it degrades for others. You have basically changed the whole dynamic. Sometimes heavy compute bound tasks are throttled, and the IO bound tasks run faster because they have access to the unused capacity. This can be good, but not so good if the compute bound tasks are the important ones. There are many greybeards, or not so grey, that remember running G3 and G4 CMOS while our mates at the pub were running Skylines. My experience is that often the CMOS with many, slower Uniprocessors would compensate for an inadequate IPS or WLM setup, throttle the Compute bound tasks, and improve the net throughput (UOW per minute). However I also remember many CICS and IDMS shops that swapped to fewer and faster Uniprocessors on Amdahl and HDS and quickly improved the throughput, and could actually use 100% of the net capacity. So the short version of all this is YMMV. You should take a bit of time to profile your workload to establish if there is a significant amount of Compute bound workload, decide how important it is to what you do, and if you will end up with reaction a, b or c. Ron > -----Original Message----- > From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of > Crispin Hugo > Sent: Friday, July 16, 2010 11:21 AM > To: IBM-MAIN@bama.ua.edu > Subject: Re: [IBM-MAIN] 2 versus 4 processors > > Hi Ron, > I am sorry but I don't understand what you mean. > ---------------------------------------------------------------------- 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
