Path: utzoo!mnetor!uunet!husc6!mit-eddie!uw-beaver!cornell!batcomputer!pyramid!mips!mash From: mash@mips.UUCP (John Mashey) Newsgroups: comp.arch Subject: Re: Horizontal pipelining [really: multi-tasking is alive and well] Message-ID: <1215@winchester.UUCP> Date: 2 Jan 88 22:02:03 GMT References: <201@PT.CS.CMU.EDU> <388@sdcjove.CAM.UNISYS.COM> <988@edge.UUCP> <1062@winchester.UUCP> <1314@sugar.UUCP> Reply-To: mash@winchester.UUCP (John Mashey) Organization: MIPS Computer Systems, Sunnyvale, CA Lines: 71 In article <1314@sugar.UUCP> peter@sugar.UUCP (Peter da Silva) writes: >In article <1062@winchester.UUCP>, mash@mips.UUCP (John Mashey) writes: >> Given that PCs were hardly the first single-user computers, >> and that multi-tasking on single-user systems has existed almost as long as >> the systems have, I'd guess that the current prevalence of >> single-tasking systems is just an anomoly of the cost/performance & >> feature (i.e. lack of builtin memory-mapping) combinations of late 70s / >> early 80s microprocessors. Peter adds some various notes to this: I don't think there is any fundamental disagreement, but I'll explain more what I meant: >I'd say that the reason was that IBM dived into the market with a machine that >was at most a slight improvement over existing single-tasking systems just as >multitasking CP/M derivitives (MP/M, CP/M 3.x) and reasonable micro-based >UNIX boxes (Onyx, etc...) were just getting into the water. IBM's entry swamped >them and they never came up again.... Apples were this way too, so one can't just blame IBM. > >> For many years, each successive generation of >> computers (mainframes, minis, micros) seemed to repeat most of the mistakes >> of the earlier ones. Now that current micros have: >> a) useful addressability (32-bits) >Multitasking doesn't require a large adress space. UNIX *used to* run fine in >256K with a user or two. Well, at one time it ran OK in 64K total, split between kernel and user, and for years, it ran OK in 124K total real memory, with 64KB I & 64KB D space per user. They were wonderful at the time, but I sure wouldn't want to go back. The compromises got pretty painful towards the end. The point is that micro addressability is in the same league as supermini and mainframe. >> b) on-chip MMUs (hence no cost-cutting reason to omit them) >Multitasking doesn't require an MMU. VenturCom has been selling UNIX for the >IBM-PC for years. Or look at the Coco... Of course it doesn't require it, but it sure makes life easier for certain styles of multi-tasking, especially UNIX's. As a percentage of the market, I suspect UNIX on PCs is a fairly low number; it is clear that 386s will make life easier. ... >> d) more-or-less reasonable design for use with multi-tasking OSs I.e., designed to support kernel + multiple user processes, each in their own address space if they want to be, allow paging, and having no oded quirks that let user programs unduly affect the state of the machine, plus enough performance that one can burn some of it supporting the extra scheduling, protection, paging, etc. [some of this has nothing to do directly with multi-tasking, maybe I should have been more specific and added some of the rest of these things, i.e., not just multi-tasking, but "reasonably modern OS that can support multiple processes, some of which might be larger than real memory, can support complex networking environments, and does not run into arbitrary small limits in physical memory, disk sizes, virtual memory, or number of processes." >Please explain what you mean by this. > >> and given that DRAMs are getting big enough that it's almost HARD to build >> tiny-memory systems, the original reasons for this aberration are rapidly >> going away, leaving only the software legacy [unfortunately]. >The original reason for this abberation is still around, but it's working hard >to catch up. Again, I don't think we fundamentally disagree, except I think blaming it all on IBM is a little strong, but that's a matter of opinion. -- -john mashey DISCLAIMER: UUCP: {ames,decwrl,prls,pyramid}!mips!mash OR mash@mips.com DDD: 408-991-0253 or 408-720-1700, x253 USPS: MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086