home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.arch
- Path: sparky!uunet!decwrl!adobe!usenet
- From: zstern@adobe.com (Zalman Stern)
- Subject: Re: Re: What's in a name? (cross-posted to SGI since mentioned)
- Message-ID: <1992Jul30.000635.11094@adobe.com>
- Sender: usenet@adobe.com (USENET NEWS)
- Organization: Adobe Systems Incorporated
- References: <58220001@sherpa2.lan1.eldec.com>
- Date: Thu, 30 Jul 1992 00:06:35 GMT
- Lines: 61
-
- In article <58220001@sherpa2.lan1.eldec.com> emagnuso@sherpa2.lan1.eldec.com
- (Erik Magnuson) writes:
- [I mention the HP3000->precision architecture and DEC VAX->Alpha
- transitions.]
- [Erik objects to my comparing systems vendors to Intel.]
-
- These are interesting examples to see how backwards compatibility can be
- dealt with. Some of the solutions used in these projects apply to the Intel
- architecture. (And in fact, Microsoft is using similar technology in their
- "Virtual DOS Machine" (VDM) for NT.) However my main point is that the
- evolution of the x86 architecture has made bows to backwards compatibility
- which have cost a lot and bought little. Fortunately, Intel has done better
- with the 486 (and probably the 586) as compared to the 286 and 386.
-
- > DEC has to support VMS, OSF/1 (aren't they dropping Ultrix?) and (in
- > the future) Windows NT. (Two of these are designed to be "portable"
- > OS's.) Systems built around the Intel line have to support MS-DOS, DR
- > DOS, Windows, OS/2, Unix, and tens or even hundreds of variations on
- > the above most of which are intimately tied to the architecture. Not
- > to mention that the price points of the system (and the total units
- > shipped) involved are an order of magnitude or two different.
-
- In both cases the goal is for *application* binaries to run reliably, with
- acceptable performance and the exact same user interface while supporting
- much better performance and functionality for future applications. In that
- vein, DEC has put significant effort into running Ultrix MIPS binaries on
- Alpha OSF/1 systems. (And apparently, if you believe Microsoft's NT spiel,
- you only need to support the top few thousand applications or so to win.)
- But once again what does this have to do with my point? How much software
- does anybody run in '386 mode on Intel processors? How much of that software
- benefits from having an instruction set closely modeled after the 8086?
-
- [My text deleted.]
- > I think this is expressed a bit backwards. Many older programs exploit
- > "features" of the previous implementations to maximize performance on
- > a limited resource machine (i.e. 8088).
- A less charitable view is that the 8086 was designed with a lack of
- foresight forcing developers to use such hacks.
-
- > These same tricks are often
- > not the fastest way to do things on a 486. What's your point? Some
- > people will always write for the lowest common denominator and some
- > will kill for an additional 10%. (Now where have I heard that before?)
- >
- > When Intel want to make existing software run faster, they introduce a
- > new part.
-
- New versions of the x86 line have not necessarily resulted in the fastest
- way to run MS-DOS programs. At the same time it has been significantly
- difficult for old software to use new architectural features. (I.e. it
- wasn't a matter of changing a few things to gain your speed back.)
-
- In other words, if you wanted to run MS-DOS applications fast when the 386
- came out, the fastest 286 around was the best way to do it. If you wanted to
- run a 32 bit OS, you couldn't run MS-DOS programs anyway except for virtual
- 8086 mode. I consider virtual 8086 mode to be a good thing to do. However,
- architecting the 32 bit instruction set to have segment registers, prefix
- instructions, and even more complex addressing modes is bogus.
- --
- Zalman Stern zalman@adobe.com (415) 962 3824
- Adobe Systems, 1585 Charleston Rd., POB 7900, Mountain View, CA 94039-7900
-