https://www-01.ibm.com/servers/resourcelink/lib03060.nsf/pages/lsprindex?OpenDocument Large System Performance Reference.
On Sat, Dec 14, 2019 at 9:29 AM Charles Mills <[email protected]> wrote: > > Responding to @Phil, yeah, don't you love it: apply a magical fudge factor > and then report the results to five digits of precision. The good news is > that both of the machines in question are the same generation: they are both > "z11's" so to speak (z196 and z114). > > Responding to @Scott, thanks, the most thorough explanation so far. > Fortunately this is not "whether an upgrade delivered the expected results?" > or "which would be better, a used z14EC or a new z15BC?" It is just "how do I > compare the CPU times of two jobs on two different machines?" Seems like a > simple question, no? > > The numbers below (from IBM.com) do not seem to support what you are saying > however: "if you're trying to convert CPU time between machines, the ratio of > any of SUs, MSUs, or PCI will be pretty much equally "fine"." The ratio of > the PCI's of the two machines is about eight-to-one but they seem in practice > to be *about* the same speed: that is, a job that uses about 1 CPU second on > one seems to use about 1 CPU second on the other (certainly not eight times > as much!). The SU/SEC ratio for the two machines is 40404/33333 which seems > to more accurately reflect observed reality (although way less than > perfectly! -- less perfectly than a guess of "oh, I guess they are about the > same speed"). > > Processor #CP PCI MSU MSUps Low Average High > 2817-730 30 23,929 2,855 2,370 49.54 42.75 37.96 > > Processor #CP PCI MSU Low Average High > 2818-Z05 5 3,139 388 6.18 5.61 > 4.77 > > In short, I'm still struggling for an answer to my simple <g> question: how > do I compare CPU times on two different machines? > > Charles > > > -----Original Message----- > From: IBM Mainframe Discussion List [mailto:[email protected]] On > Behalf Of Scott Chapman > Sent: Saturday, December 14, 2019 4:48 AM > To: [email protected] > Subject: Re: How do I compare CPU times on two machines? > > SUs, MSU, PCI (IBM MIPS) ratings are all just different magnitudes of the > same number. What I mean is that they all are calculated from the same LSPR > tests and exist in relatively fixed ratios to each other. There may be some > slight variations because (for example) MSUs and PCIs are quoted in whole > numbers and IBM seems to tweak them very slightly. The ratios differ slightly > between single and dual frame, full speed and sub-cap engines, and number of > engines. But for all practical purposes, the ratios hold within a couple > percent. > > Re. the "technology dividend" where they derated the MSUs relative to the SUs > (before they decided to deliver software price improvements by software price > changes), that only changed the ratios between MSUs and the SUs. > > In the old days (before the "technology dividend") MSUs ~ SUs * CPUs * 3600 > Now, for the last several generations of machines: MSUs ~ SUs * CPUs * 3600 * > 0.664 > > With the z15, the ratio between SUs and PCI are even flatter than they were > in prior generation. I.E. my understanding is that they're doing much more of > a straight-up calculation and not "tweaking" (my term) the results as much as > they did prior. (And to be clear: the prior variation due to "tweaking" was > not very much, it's just less with the z15.) > > So in short, if you're trying to convert CPU time between machines, the ratio > of any of SUs, MSUs, or PCI will be pretty much equally "fine". Not > necessarily accurate, but all of them will be about the same. > > If you want to be more accurate about it (such as evaluating whether an > upgrade delivered the expected results), then build zPCR models of the two > machines in question and use the ratios that it produces. But in all cases, > the expectation is that the ratio between the two machines is an average of > many different types of work. Individual work units will over- or > under-perform expectations. The hope (and real expectation) is that across > all the work on the system you come close (+/- 5%) to the ratio provided by > zPCR. Reality may differ more significantly from expectation if you're just > using one of the single-number metrics without regard to the RNI of the work > and the LPAR configuration (factors that zPCR takes into account). > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to [email protected] with the message: INFO IBM-MAIN -- Mike A Schwab, Springfield IL USA Where do Forest Rangers go to get away from it all? ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
