the i860 and the 432 (was Re: Squeak History / Tiny Machines)

Alan Kay Alan.Kay at squeakland.org
Wed Mar 19 04:45:04 UTC 2003


Just curious. One of the many flaws in the 432 was that it was a 
"large object" architecture that IIRC was derived from the Hydra OS 
done at CMU. So it treated "small objects" not as objects but as 
data: numbers, strings, etc., and missed the B5000 or especially 
Smalltalk architecture that could have been used. Did they fix this 
in the succeeding chip designs?

Cheers,

Alan

At 8:07 PM -0800 3/18/03, Randal L. Schwartz wrote:
>  >>>>> "Allen" == Allen Wirfs-Brock 
><Allen_Wirfs-Brock at Instantiations.com> writes:
>
>Allen> Actually, I believe that they did. However, it also
>Allen> failed. Intel form a joint venture with (I think) Siemens call
>Allen> Biin that was essentially a follow on to the 432 work.  My
>Allen> understanding is that the i860 processor family came out of
>Allen> that effort after Biin, itself was aborted.  The i860 was a
>Allen> RISC based processor that ended up being mostly used in
>Allen> embedded applications.  However, if you looked closely at the
>Allen> architecture there were strange and interesting features that,
>Allen> if enabled, would be useful in building a capability based
>Allen> system.
>
>The i860 was a family of processors.  The KA was simple integer math,
>the KB added floating point, the MC added military-grade compatibility
>and the XA (the chip used by BiiN) was the 432-reborn (hence the code
>name "phoenix" before BiiN).
>
>The XA had hardware support as an Object Machine, with opaque object
>handles and full inheritance and capabilities.  It was quite nice and
>secure.  A class could return an instance, and you couldn't even look
>inside that instance... all you could do is call methods which called
>back to the class.
>
>I wrote the Intel i860XA Architecture Manual (essentially the chip
>spec).  The doc project was over budget and behind schedule... I was
>called in, scrapped everything I saw, started from scratch, and hit
>the original deadline!  (My OO experience far exceeded the previous
>techwriter, so I actually "got" the purpose of the chip, and that
>helped.) I still have a copy of the XA manual here on my sample shelf
>in my office.
>
>Now for the secret that can be told a decade later.  The initial 860
>series from KA to XA were all the same chip on one die.  Every
>function was tested completely on every chip, and then in the final
>stage a different label was put on the chip, varying the price from
>$30 to $350. :) Eventually, there was die differentation, but I found
>it amusing that someone could pay an order of magnitude more simply
>for a label change.
>
>And for another twist of fate, I was also employed by Sequent
>(employee #64), which was started by Casey Powell after he walked out
>of the Intel 432 project muttering "we should do this thing with
>commodity hardware instead of inventing everything from scratch".  So
>I worked on *both* major derivatives of the failed 432 project. :)
>
>--
>Randal L. Schwartz - Stonehenge Consulting Services, Inc. - +1 503 777 0095
><merlyn at stonehenge.com> <URL:http://www.stonehenge.com/merlyn/>
>Perl/Unix/security consulting, Technical writing, Comedy, etc. etc.
>See PerlTraining.Stonehenge.com for onsite and open-enrollment Perl training!


-- 



More information about the Squeak-dev mailing list