The following message is a courtesy copy of an article that has been posted to bit.listserv.ibm-main,alt.folklore.computers as well.
[EMAIL PROTECTED] (, IBM Mainframe Discussion List) writes: > I made a mistake. A track not in the cache would take on the order of 20 > milliseconds, so that would equate to 20 days instead of one day. A track > already cached would result in an access time of one millisecond. If the 4K > block can be found in a buffer somewhere in virtual storage inside the > processor, > it might take from 100 to 1000 instructions to find and access that data, > which would equate to 100 to 1000 seconds, or roughly one to 17 minutes. > And > that assumes that the page containing the 4K block of data can be accessed > without a page fault resulting in a page-in operation (another I/O), in > which > case we are back to several days to do the I/O. > > By the way, it takes at least 5000 instructions in z/OS to start and finish > one I/O operation, so you can add about two hours of overhead to perform the > I/O that lasts for 20 days. > > You really want to avoid doing an I/O if at all possible. reply to comment about RPS-miss (in the vmesa-l flavor of this thread) http://www.garlic.com/~lynn/2007s.html#5 Poster of computer hardware events? i had been making comments over a period of yrs that disk relative system thruput had declined by an order of magnitude (i.e. disks were getting faster but processors were getting much faster, faster). this eventually led to somebody in the disk division (gpd) to assigning the gpd performance group to refute the statements. after several weeks they came back and effectively said that i had somewhat understated the disk relative system thruput degradation ... when RPS-miss was taken into account. they then put a somewhat more positive spin on it and turned it into share 63 presentation b874 ... some past references: http://www.garlic.com/~lynn/2002i.html#18 AS/400 and MVS - clarification please http://www.garlic.com/~lynn/2006f.html#3 using 3390 mod-9s http://www.garlic.com/~lynn/2006o.html#68 DASD Response Time (on antique 3390?) one of the issues is does the 5k instruction pathlength roundtrip from EXCP (including channel program translation overhead) or roundtrip just after it has been passed to i/o supervisor??? for comparison numbers ... i had gotten cp67 total "roundtrip" for page fault down to approx. 500 instructions ... this included page fault handling, page replacement algorithm, a prorated fraction of page i/o write pathlength (which includes everything to start/finish i/o), total page i/o read pathlength (including full i/o supervisor), and two task switches thru dispatcher (one to switch to somebody else, waiting on the page fault to finish and another to switch back after the page i/o read finishes). to get it to 500 instructions involved touch almost every piece of code involved in all of the operations. I believe the "5000" instruction number was one of the reasons that 3090 extended store was a syncronous instruction (since the asyncronous overhead and all related gorp in mvs was so large). earlier, there had been some number of "electronic" 2305 paging device deployed at internal datacenters ... referred to as "1655" model (from an outside vendor). these involved effectively low latency but limited to channel transfer and cost whatever the asyncronous processing overhead. the 3090 extended store was done because of physical packaging issues ... but later when physical packaging was no longer an issue ... there were periodic discussions about configuring portions of regular memory as simulated extended store ... to compensate for various shortcomings in page replacement algorithms. with regard to the cp67 "500" instruction number vis-a-vis MVS ... i would periodically take some heat regarding MVS having much more robust error recovery as part of the 5000 number (even tho the 500 number was doing significantly more). so later when i was getting to play in bldgs 14 & 15 (dasd engineering lab and dasd product test lab), i had opportunity to rewrite vm370 i/o supervisor. the labs in bldg. 14&15 were running processor "stand-alone" testing for the dasd/controller "testcells" (one at a time). They had tried doing this under MVS but had experienced 15min MTBF (system crashing and/or hanging with just a single testcell). I undertook to completely rewrite i/o supervisor to make it absolutely bullet proof, allow concurrent testcell operation in operating system environment. lots of past posts mentioning getting to play disk engineer http://www.garlic.com/~lynn/subtopic.html#disk some old postings about comparisons of degradation of disk relative system thruput. the claim was that doing similar type of cms workload ... in going from cp67 on 360/67 with 80 users to vm370 on 3081 ... it should have shown an increase to several thousand online uses ... instead of increase to 300 or so online users. The increase in online users is roughly the change in disk system thruput ... as opposed to difference in processor thruput http://www.garlic.com/~lynn/93.html#31 Big I/O or Kicking the Mainframe out the Door http://www.garlic.com/~lynn/94.html#43 Bloat, elegance, simplicity and other irrelevant concepts http://www.garlic.com/~lynn/94.html#55 How Do the Old Mainframes Compare to Today's Micros? http://www.garlic.com/~lynn/95.html#10 Virtual Memory (A return to the past?) http://www.garlic.com/~lynn/98.html#46 The god old days(???) http://www.garlic.com/~lynn/99.html#4 IBM S/360 http://www.garlic.com/~lynn/2001d.html#66 Pentium 4 Prefetch engine? http://www.garlic.com/~lynn/2001f.html#62 any 70's era supercomputers that ran as slow as today's supercomputers? http://www.garlic.com/~lynn/2001l.html#40 MVS History (all parts) http://www.garlic.com/~lynn/2001l.html#61 MVS History (all parts) http://www.garlic.com/~lynn/2001m.html#23 Smallest Storage Capacity Hard Disk? http://www.garlic.com/~lynn/2002.html#5 index searching http://www.garlic.com/~lynn/2002b.html#11 Microcode? (& index searching) http://www.garlic.com/~lynn/2002b.html#20 index searching http://www.garlic.com/~lynn/2002e.html#8 What are some impressive page rates? http://www.garlic.com/~lynn/2002e.html#9 What are some impressive page rates? http://www.garlic.com/~lynn/2004p.html#39 100% CPU is not always bad ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html