Sun 386i (WAS: RE: [rescue] Interactive Unix?)

[Greg A. Woods]

[ On Tuesday, December 11, 2001 at 15:45:33 (-0500), Corda Albert J DLVA wrote: ]
> Subject: RE: Sun 386i (WAS: RE: [rescue] Interactive Unix?)
>
> Yes, It's ages shows... but you have to remember that at the
> time of it's inception, the 8086 architecture made sense.

I beg to differ.  It was already 5 years behind the times before it went
into early sampling.

> Very few
> microprocessors had companion MMU chips, and virtually no
> microprocessor had an MMU integrated into the CPU chip itself.
> The segmented architecture provided a cheap and efficient way to
> generate pseudo-relocatable code with a minimum of hassle.

Having a separate MMU wasn't the issue -- the 8086 didn't even make it
easy to have one!  The only reason it made sense was that some idiots
thought the world would be happy with the much older software that would
still run on it.

Of course Motorola wasn't much more advanced -- they were stumbling
trying to get a real computer working and failed miserably with the
68010.  However with the 020 they blew the doors off Intel, and even
with lower sales and less marketing they still managed to support
development of processors at least a generation ahead of Intel right up
to the 060.

Intel was a bunch of embedded systems weenies with too many military
contracts to know what was needed in the real world.  If they'd taken
even a peek at the VAX or heaven forbid a real mainframe and if they'd
taken the time to have a systems programmer understand their needs

Of course even when you get everything done in-house, eg. like the MIPS
or SPARC, you still end up with hardware weenies not listening to what
their real requirements are for efficient systems programming.  Larry
McVoy gave a talk at a Usenix conference a few years ago lamenting at
how hard it was to try to knock some sense into hardware engineers
working at the same company he was working at.  That was the talk about
lmbench, and he said as much that even with the raw numbers from lmbench
staring them in the face they still had trouble understanding what
software people really wanted from them.

> I remember an experiment/development demo board kit that was
> marketed in an Intel catalog of that era which, with a
> minimum of chips (only an 8088, some glue logic, and memory)
> allowed you to build a "multi-user" basic (as in
> basic-the-langage) system. This was a marvel at the time to
> those of us who were building systems at the chip level.

That was a bold-faced marketing lie.  I remember looking at the
preliminary specs for that thing and being somewhat impressed, and then
later looking at the final specs and wondering what the hell I'd been
thinking earlier.  Even with the 286 it still wasn't possible to build a
really proper multi-user system with even basic process swapping (which
is why Xenix/286 sucked so much).

> In
> comparison to the other CPU chips of that era, it's hardware
> architecture was pretty advanced.

It was technically advanced in terms fo chip design, maybe.

But as a processor they probably could have gotten twice as much out of
it with the same number of transistors......

> Where it bit the proverbial
> rag was in it's insistance on sticking with a dedicated-register
> architecture in order to try to recapture some of the huge
> existing base of 8085 code (although it was not binary-compatible,
> intel made a big deal about providing "source-code translators").
> The segmented model made it easer to build a 16-bit CPU that
> looked somewhat like the 8085 architecture that everyone was
> familiar with. Unfortunately, I don't think anyone at that
> time either really fully appreciated or cared about the dreaded
> "backwards compatibility" software demon that still nags
> us to this day.

Like I said -- all Intel hardware heads were stuck deeply into the
embedded systems clouds with no clue about where people were really
trying to use microprocessors in business applications.

> All in all, I kinda wish that the 68000 had won the CPU "wars"
> of that era. It was an altogether "kinder and gentler"
> software architecture, and had the same kind of advanced
> hardware architecture (i.e. separation of the CPU core from
> what Intel called the BIU (Bus Interface Unit), which allowed
> both of them to build chips with both an external 8-bit bus
> (8088 and 68008) and 16-bit bus (8086 and 68000)... but
> that's another story :-)

If National could have afforded to hire a few more more good silicon
engineers and quality assurance types they would have had the best chip
going.  The first ns32032 system I did some simple tests on blew the
doors off a mc68020 system with twice as much RAM in it.

Even now Intel's only gained some respectability by buying all of
digital's chip fabs and engineering talent.  Little good it's done the
rest of us.

-- 
								Greg A. Woods

+1 416 218-0098;  <gwoods at acm.org>;  <g.a.woods at ieee.org>;  <woods at robohack.ca>
Planix, Inc. <woods at planix.com>; VE3TCP; Secrets of the Weird <woods at weird.com>