wilson@carcoar.Stanford.EDU (Paul Wilson) (09/09/89)
In article <45344@bbn.COM> slackey@BBN.COM (Stan Lackey) writes: >In article <2089@uceng.UC.EDU> dmocsny@uceng.UC.EDU (daniel mocsny) writes: >>In article <278@baird.cs.strath.ac.uk>, jim@cs.strath.ac.uk (Jim Reid) writes: >>> A VM system general enough to perform well for most potential users and >>> their applications on most potential hardware platforms is asking a lot. >> >>And the cost of failing to give this to the users is a lot more. > >Giving up something is required for generality; no general purpose >machine is going to run FFT's as fast as an FFT box that costs the >same. > >-Stan Is this so? My impression is that UNIX systems don't generally use things like Working Set schedulers because nobody has bothered to implement them, not because they sacrifice generality. Working Set, as I understand it, is simply superior in performance to something like a global LRU, and cheap approximations to it aren't that hard to implement. So in fact a good approximation of Working Set is *more* general than what most people actually use. In the easy cases it works about the same, or a bit better. And in the hard cases it works a *lot* better. [It may be that UNIX doesn't allow users to specify enough about the priorities of jobs to enable a WS scheduler to do a lot of smart process-swapping, but a WS page-replacement algorithm should still be a win and reduce page faults. And extensions to UNIX to support WS process scheduling wouldn't hurt the people who didn't use them] I'm sure somebody will correct me if I'm wrong, but I though good approximations to WS were achievable on just about any paged hardware, even lacking reference bits, at a small cost. I think there was a paper on something called WSClock, showing how to do this, by somebody or other and Hennessy, in the 1984 (83?) SOSP. So now I'm curious -- is it just that most UNIX implementors haven't gotten around to using this technology, because it's not an issue for most of their users, or are the supposedly cheap approximations oversold? (I've added a couple of OS newsgroups, to hear what OS people think, but followups are to comp.arch.) -- Paul Paul R. Wilson Software Systems Laboratory lab ph.: (312) 996-9216 U. of Illin. at C. EECS Dept. (M/C 154) wilson@carcoar.stanford.edu Box 4348 Chicago,IL 60680