Anne & Lynn Wheeler wrote:
I even had something of a set-to with the POK performance modeling group on the subject during the early days of SVS development ... and some choice they made for page replacement ... which took well into MVS releases ... before they could understand how really bad the design choice was.
... some old email ... From: wheeler Date: 05/06/81 14:41:17 re: unchanged pages on the free list 1st; I've been repeating that you can't screw up a LRU algorithm that way. I told the AOS people that way back in 71-72 but they wouldn't listen. If you are going to permute the LRU algorithm you have to do it randomly. Unchanged pages is a non-random permutation of the LRU algorithm. AOS went 1st customer ship & a couple of years before they realized that shared, nucleus storage was being biased against in favor of private, individual data areas (i.e. pages in shared LINKPACK would be biased against, in favor of private data pages). A LRU algorithm, is a LRU algorithm, is a LRU algorithm. There is at least as good a case or better for selecting change pages to be placed on the free list ahead of non-changed pages (there is higher probability that r/o pages are program pages, there are lot more instances of re-use of program pages while data pages change than there is of the inverse). Problem of concentrating/optimization for just one area of the system w/o paying attention to overall system relationships & design can lead to unatticipated problems & results. Page replacement is a total complex system problem which spans the use of CPU, real storage and I/O deivces. Local optimization very quickly can lead to sub-optimal global system optimization. ... snip ... in the above, "AOS" was the prototype of VS2/SVS ... which morphed into VS2/MVS ... still with the same replacement algorithm. It wasn't until well into the MVS release cycle that they understood how bad it was and changed it. note, "is a LRU algorithm" ... there is, of course, "local LRU" and "global LRU". I had done "global LRU" as an undergraduate in the 60s ... at about the same time that there was some amount of academic literature on "local LRU". More than a decade later there was a lot of uproar generated about awarding a Stanford Phd on the subject of "global LRU". recent post http://www.garlic.com/~lynn/2006w.html#45 The Future of CPUs: What's After Multi-Core? with some old communication (19Oct82) jumping right into the middle of dispute (took me nearly a year to get approval to send the communication): http://www.garlic.com/~lynn/2006w.html#email821019 misc. other "old" email on the subject of "global LRU" http://www.garlic.com/~lynn/lhwemail.html#globallru and collected posts mentioning the subject http://www.garlic.com/~lynn/subtopic.html#wsclock ---------------------------------------------------------------------- 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
