[cap-talk] tagged architectures

[Jed at Webstart]

At 09:46 AM 11/10/2004, Michal Kostal wrote:
>Hello,
>     in the current time I am searching for the computer architectures
>which efficiently support the secure computing environments. I founded
>out that in 70's and 80's there were high expectations in the use of
>tagged architectures, but from the middle of the 80's all efforts
>about it suddenly stopped.
>     Does anybody know if there are some recent architectures based on such
>design, or if there are some articles which describes why is it no more
>attractive? Your opinions about it would help me, too.

I'm sure opinions will vary, but I'll share mine.  In the 70s and early 80s
there was much more variety in general in hardware (computer architecture,
networking, peripherals in general) and software (operating systems
in particular).  It was a time when there was no clearly dominant
and successful paradigm.  Once the microprocessor revolution
took off and became dominated by just a few companies (Intel,
AMD, and IBM pretty much now) and home computers became
such a large market with the likes of Microsoft dominating in
the software area there just wasn't (isn't) much room for such variety.
Distinctive architectures just can't get any traction.  Systems are
assumed to have a much larger and more capable set of support
software on them these days.  The cost of getting a new architecture
to a point where it has comparable support software is high enough
that any would be innovations are pretty much snuffed out.  Also,
what's the motivation for any such innovations?  Typically they
show up on academic or research areas at least first.  However,
the academic and research organizations typically can't afford
work in the hardware area.

One other factor that I think may have had a significant influence
in the 70's (at least) and perhaps into the 80s was the prevalence
of micro programmed architectures.  This concept is largely
unknown today with the dominance of microprocessors, but in
the 70s it was a (the most?) common architecture for processors.
With a micro programmed architecture it was much easier to
vary things like tagging with relatively low cost changes in the
microcode.

Another minor factor may have been the emergence of efficient,
relatively low cost and byte/word oriented solid state memories.
With such memories it's difficult to allocate a tag field without
chewing into an accepted (power of 2) word size.

Having said the above, I think that while there was a good deal
of experimenting with tagged architectures in the 70s and some
into the early 80s (IBM), the case for such architectures seemed to
be weak.  None of them (in my opinion) seemed to demonstrate
a clear advantage from the tagging.  I have perhaps most experience
with the Burroughs systems (B5000, B6000).  Those were very much
Algol systems.  They did seemingly amazing things with tagging
where a word could be a:

array descriptor (indexed or not)
indirect reference word
program control word
(others - I don't remember off hand)

and any reference to a variable could automatically do things
like redirect, index into an array (even indexing from the stack),
or even enter a program (subroutine) and pick up with the returned
value that was left on the stack.

Still, for all such things (and perhaps even for what to me seem
more reasonable efforts in the security area) there was no clear
demonstration of value beyond some efficiency that was more
than made up for in newer and simpler architectures that weren't
tagged and had wider commercial appeal.

If you couple the above together it seems quite clear to me
why there have been no tagged architectures after the early
1980s.

--Jed http://www.webstart.com/jed/