[net.micro.68k] 68000 Based UNIX Summary

bj (08/18/82)

Several people have complained that they did not see this article
the first time around, so here it is again:

I received several responses to my request for 68000 UNIX information.
The most important was a summary from aron shtull-trauring
(harpo!presby!aron) of the information he had received from his earlier
request.  The most significant new information is:

----------------------------------------------------------------------
    Consider this contribution annonymous:

            Radio Shack will be supplying what you want.  The target
    date is late November.
----------------------------------------------------------------------

MIT, Stanford, Lucasfilm, SMI, Cadlink, and most likely
all the other SUN-based workstation companies all
offer some version or another of V7 UNIX for the 68000.

************

Unsubstantiated rumor claims that Microsoft is doing their usual half-assed
job bringing up Unix on the Tandy machine.  The TRS-80 mod 16's internal
architecture is pretty awful so even a good Unix might not run all that
well.


************  [The rest of this message is the previous summary.  B.J.]


As has been mentioned in WorkS, the 68000 doesn't really support paging
or even the automatic stack growth that UNIX uses.  So a *good* UNIX isn't
likely till the 68010 or 68020 are available.

The major problem is that an instruction that is aborted in the middle
by any sort of memory management fault cannot in general be restarted;
thus, demand paging is totally impossible.  On 11s, that isn't a problem,
but UNIX does depend on page faults on stack references to know when to
expand it.  This could be faked by including some restartable (or perhaps
unneeded) instruction at the entry to each block that would reference the
bottom of the new stack; that could force the expansion.  Of course, there
would be extra overhead for each subroutine call in that case.

Microsoft has committed to producing UNIX versions for all of the major
16-bit micros, including the 68000.  To the best of my knowledge, the
only one they're marketing now is for LSI-11/23s.  I strongly suspect
that they will wait until they can get a UNIX 3.0 -based system before
releasing any others.  Not that it's necessarily technically better, but
Bell's licensing terms (for them) are far superior.

The real problem, though, will be with the disk -- UNIX is a heavily
I/O-bound system, and can't really survive on floppies.  A 5-10M
Winchester is the minimum configuration, I'd say, that will really
support UNIX.

Several folks other than Microsoft are also planning "UNIX-like" systems.
I am told that HP will offer one; I don't know, though, if they're doing
it themselves or what.

************

     If you read WorkS digest you should have seen an entry about the
     Sun workstation. It is a 68000 based graphics work station and it
     supports Unix. You may be able to get the software for a TRS-80
     from the people at Stanford who are working on it.

************

Gee -- after saying it would be a while, I just saw an ad in the Feb. BYTE
for a 68000-based V7 UNIX system.  The ad is by "DUAL" -- no mention of
hard disks, but they list a beginning price of $8295.  The UNIX system is
by UNISOFT of Berkeley, CA.


************

        In Thomas/Yates (A User Guide to the Unix System, Publ. by Osborne-
McGraw-Hill in Berkeley) the following are listed:

Unix:
        (1) Microsoft, 400 108 Av., N.E., #200, Bellvue, WA 98004
                (XENIX--a Unix look-alike)
        (2) Uni-Soft, 2405 4 St., Berkeley, CA 94710
        (3) Mark Williams Co., 1430 W. Wrightwood Av., Chicago, IL 60614
                (COHERENT, an O/S based on Unix)
        (4) Whitesmiths, Ltd., 1780 B'way, #601, New York, NY 10019
                (IDRIS, an O/S based on Unix)
C:
        (1) Alcyon Corp., 8474 Commerce Ave., San Diego, CA 92121

        I'm certain there are many others.  I know Lucasfilms, (the Star
Wars Company!) has brought a 68000 Unix, but isn't supporting it; they
should be listed in the Marin County (area code 415) phone book.

************

UNIX on 68K... originally done at MIT, available from LUCASFILM,
(on a SUN MULTI-bus system), WICAT on their own 68K system, Microsoft
(XENIX) on a QBUS 68K board--soon on other processors, Fortune on their
own system (proprietary bus), plus a few more in progress...

************

Whitesmith's and lots of other people have 68000 based UNIX products.
There must have been almost a dozen vendors of that type of
gear at last week's UNIX conference.  Read BYTE for more
info on this kind of things.
                          ^ thing, not things.  Damn mail program.

************

*Unix has been brought up on a 68000. What I was suggesting was that if
 one was going to buy a Sun it would be better to wait for the
 68010 since demand paging gives one greater capabilities with respect
 to timesharing and having multiple processes running simultaneously.
 But 68000 Unix is up and running well from what I can discern.

************

Yes, the 3.0 license will allow cheaper Unices, but I suspect that most
vendors haven't converted yet; given how slowly Western moves, they may
not even have their licenses yet.  I'd give it a few more months.  I've
been told that a 3.0 license subsumes v7 and 32V, so presumably anyone
who's now marketing a v7-based system will be able to sell the same thing
under the new terms, but I'm not certain of that.

About Xenix:  yes, it is real UNIX v7.  They've rewritten some sections
for improved reliability and speed, and added some stuff, but as far
as I know, any program that runs under v7 should run just fine under it.
Based on my own experiences in transporting UNIX to other machines, I'd
guess that they're also cleaning up and reorganizing code some, so that
they can support multiple machines with just one set of common source code.
For example, very many programs have a hidden dependency on the sign bit
of char variables being propogated when assigned to an int; this causes
them to break on most non-DEC machines.


************

Well, here's what I could find out so far.

There are alot of people coming out with bundled hardware/software
systems involving 68000's and Unix.  Among these are WICAT, MicroDaSys,
Dual and Fortune.  It seems every week that there is an article in
INFOWORLD announcing a new system of this sort.  In fact, in this
weeks INFOWORLD there is an article on Unix and 16 bit micros.  Most
of the above companies have full page adds in this months BYTE.

The problem, however, for the 16, is that these systems are bundled.
The hardware and software come together.  Most of these companies
are getting Unix from two places (on an OEM basis): UNISOFT in
Berkeley, CA and Microsoft in Bellevue, WA.

UNISOFT is currently V7 but will soon be Version III (or whatever
WE calls its latest concoction).  Even nicer, UNISOFT includes
the Berkeley Unix enhancements.  MICROSOFT is also V7 and I believe
will also soon be III, but does not include Berkeley stuff, although
it does have "internals" enhancements of its own.

Neither MICROSOFT nor UNISOFT is interested in selling single user
copies.  They'll only talk to you if you are an OEM.  On the other
hand, the OEM's will only sell you Unix for a 68000 inside their
own hardware.  According to the INFOWORLD article, the new WE
licensing arrangement will allow resale prices of $40 a copy for
dealers who've shelled out the $45,000 for the source-code license.

At those prices its understandable that dealers are relunctant to
sell Unix without the hardware.  There's no profit in it!

Anyway, unless someone out there has alot of money and time, and
is willing to become a TANDY and UNISOFT(or MICROSOFT) OEM, or
unless TANDY pleasantly surprises me, it may be a while before
we see Unix on the TRS-80 16.

************

Xenix 3.0 will be available for the 68000.  Xenix 2.2 (V7) is currently
ported to the 68000, but none of the machines are ready to be shipped
(one, in fact, will never be retailed to the public).  We're doing
a version for a SUN board (very sexy!) which should be complete in
6 weeks, on the streets in another 6 or 8, however fast the OEM
moves.  The second factor is the timing of Xenix 3.0, since we want
to combine some elements of Berkeley 4.1 in the release.  I expect
it will be third quarter before XENIX 3.0 is available.  Thus, 68000s
will be available sooner, but with XENIX 2.3.

As for the TRS 68000 machine, it has no memory management in its
current form.  We probably won't do anything with it until a version
with an MMU is available.

I dont understand your question about 'paging restrictions'.  If you
mean virtual memory, we will not support that until we have a real
virtual chip, either the 68010 or the 16032.  One of our OEMs,
MicroDaSys, is porting their own kernal to a 68000, and are doing
a dual-cpu virtual implimentation, where a second CPU wakes up on
a page fault, brings in the page, and lets the second one resume
from its amazingly long memory-access wait state.  I think that this
is a highly specialized system, as it does not provide general
virtual memory performance.

If you mean the fearsome restrictions in the 68000's MMU archetecture
(the Motorolla MMU chip), we've lost a lot of sleep over that design,
and we believe that we have some approaches and techniques to handle
the problem.

As for cost, the price is set by the OEM.  Since our price to them
is quite low, I would expect the OEMs to offer a very attractive price.
Many of them plan on a nice low price for the basic system, with
additional charges for some items which they hold back, such as
(for example) yacc/lex, screen editor, etc.  Our price to the
OEM is very close to the Bell price to us, which was published by
Bell at the last Usenix.  For a 15 user system, its around $300 or
so, roughly.

                                        B.J.
                                        Herbison-BJ@Yale
                                        decvax!yale-comix!herbison-bj

Anonymous (12/06/82)

#R:yale-com:-80800:tekcrd:12600001:000:10087
tekcrd!Anonymous    Dec  6 12:56:00 1982

***** tekcrd:net.micro.68k / yale-com!bj /  4:38 pm  Aug 18, 1982
Several people have complained that they did not see this article
the first time around, so here it is again:

I received several responses to my request for 68000 UNIX information.
The most important was a summary from aron shtull-trauring
(harpo!presby!aron) of the information he had received from his earlier
request.  The most significant new information is:

----------------------------------------------------------------------
    Consider this contribution annonymous:

            Radio Shack will be supplying what you want.  The target
    date is late November.
----------------------------------------------------------------------

MIT, Stanford, Lucasfilm, SMI, Cadlink, and most likely
all the other SUN-based workstation companies all
offer some version or another of V7 UNIX for the 68000.

************

Unsubstantiated rumor claims that Microsoft is doing their usual half-assed
job bringing up Unix on the Tandy machine.  The TRS-80 mod 16's internal
architecture is pretty awful so even a good Unix might not run all that
well.


************  [The rest of this message is the previous summary.  B.J.]


As has been mentioned in WorkS, the 68000 doesn't really support paging
or even the automatic stack growth that UNIX uses.  So a *good* UNIX isn't
likely till the 68010 or 68020 are available.

The major problem is that an instruction that is aborted in the middle
by any sort of memory management fault cannot in general be restarted;
thus, demand paging is totally impossible.  On 11s, that isn't a problem,
but UNIX does depend on page faults on stack references to know when to
expand it.  This could be faked by including some restartable (or perhaps
unneeded) instruction at the entry to each block that would reference the
bottom of the new stack; that could force the expansion.  Of course, there
would be extra overhead for each subroutine call in that case.

Microsoft has committed to producing UNIX versions for all of the major
16-bit micros, including the 68000.  To the best of my knowledge, the
only one they're marketing now is for LSI-11/23s.  I strongly suspect
that they will wait until they can get a UNIX 3.0 -based system before
releasing any others.  Not that it's necessarily technically better, but
Bell's licensing terms (for them) are far superior.

The real problem, though, will be with the disk -- UNIX is a heavily
I/O-bound system, and can't really survive on floppies.  A 5-10M
Winchester is the minimum configuration, I'd say, that will really
support UNIX.

Several folks other than Microsoft are also planning "UNIX-like" systems.
I am told that HP will offer one; I don't know, though, if they're doing
it themselves or what.

************

     If you read WorkS digest you should have seen an entry about the
     Sun workstation. It is a 68000 based graphics work station and it
     supports Unix. You may be able to get the software for a TRS-80
     from the people at Stanford who are working on it.

************

Gee -- after saying it would be a while, I just saw an ad in the Feb. BYTE
for a 68000-based V7 UNIX system.  The ad is by "DUAL" -- no mention of
hard disks, but they list a beginning price of $8295.  The UNIX system is
by UNISOFT of Berkeley, CA.


************

        In Thomas/Yates (A User Guide to the Unix System, Publ. by Osborne-
McGraw-Hill in Berkeley) the following are listed:

Unix:
        (1) Microsoft, 400 108 Av., N.E., #200, Bellvue, WA 98004
                (XENIX--a Unix look-alike)
        (2) Uni-Soft, 2405 4 St., Berkeley, CA 94710
        (3) Mark Williams Co., 1430 W. Wrightwood Av., Chicago, IL 60614
                (COHERENT, an O/S based on Unix)
        (4) Whitesmiths, Ltd., 1780 B'way, #601, New York, NY 10019
                (IDRIS, an O/S based on Unix)
C:
        (1) Alcyon Corp., 8474 Commerce Ave., San Diego, CA 92121

        I'm certain there are many others.  I know Lucasfilms, (the Star
Wars Company!) has brought a 68000 Unix, but isn't supporting it; they
should be listed in the Marin County (area code 415) phone book.

************

UNIX on 68K... originally done at MIT, available from LUCASFILM,
(on a SUN MULTI-bus system), WICAT on their own 68K system, Microsoft
(XENIX) on a QBUS 68K board--soon on other processors, Fortune on their
own system (proprietary bus), plus a few more in progress...

************

Whitesmith's and lots of other people have 68000 based UNIX products.
There must have been almost a dozen vendors of that type of
gear at last week's UNIX conference.  Read BYTE for more
info on this kind of things.
                          ^ thing, not things.  Damn mail program.

************

*Unix has been brought up on a 68000. What I was suggesting was that if
 one was going to buy a Sun it would be better to wait for the
 68010 since demand paging gives one greater capabilities with respect
 to timesharing and having multiple processes running simultaneously.
 But 68000 Unix is up and running well from what I can discern.

************

Yes, the 3.0 license will allow cheaper Unices, but I suspect that most
vendors haven't converted yet; given how slowly Western moves, they may
not even have their licenses yet.  I'd give it a few more months.  I've
been told that a 3.0 license subsumes v7 and 32V, so presumably anyone
who's now marketing a v7-based system will be able to sell the same thing
under the new terms, but I'm not certain of that.

About Xenix:  yes, it is real UNIX v7.  They've rewritten some sections
for improved reliability and speed, and added some stuff, but as far
as I know, any program that runs under v7 should run just fine under it.
Based on my own experiences in transporting UNIX to other machines, I'd
guess that they're also cleaning up and reorganizing code some, so that
they can support multiple machines with just one set of common source code.
For example, very many programs have a hidden dependency on the sign bit
of char variables being propogated when assigned to an int; this causes
them to break on most non-DEC machines.


************

Well, here's what I could find out so far.

There are alot of people coming out with bundled hardware/software
systems involving 68000's and Unix.  Among these are WICAT, MicroDaSys,
Dual and Fortune.  It seems every week that there is an article in
INFOWORLD announcing a new system of this sort.  In fact, in this
weeks INFOWORLD there is an article on Unix and 16 bit micros.  Most
of the above companies have full page adds in this months BYTE.

The problem, however, for the 16, is that these systems are bundled.
The hardware and software come together.  Most of these companies
are getting Unix from two places (on an OEM basis): UNISOFT in
Berkeley, CA and Microsoft in Bellevue, WA.

UNISOFT is currently V7 but will soon be Version III (or whatever
WE calls its latest concoction).  Even nicer, UNISOFT includes
the Berkeley Unix enhancements.  MICROSOFT is also V7 and I believe
will also soon be III, but does not include Berkeley stuff, although
it does have "internals" enhancements of its own.

Neither MICROSOFT nor UNISOFT is interested in selling single user
copies.  They'll only talk to you if you are an OEM.  On the other
hand, the OEM's will only sell you Unix for a 68000 inside their
own hardware.  According to the INFOWORLD article, the new WE
licensing arrangement will allow resale prices of $40 a copy for
dealers who've shelled out the $45,000 for the source-code license.

At those prices its understandable that dealers are relunctant to
sell Unix without the hardware.  There's no profit in it!

Anyway, unless someone out there has alot of money and time, and
is willing to become a TANDY and UNISOFT(or MICROSOFT) OEM, or
unless TANDY pleasantly surprises me, it may be a while before
we see Unix on the TRS-80 16.

************

Xenix 3.0 will be available for the 68000.  Xenix 2.2 (V7) is currently
ported to the 68000, but none of the machines are ready to be shipped
(one, in fact, will never be retailed to the public).  We're doing
a version for a SUN board (very sexy!) which should be complete in
6 weeks, on the streets in another 6 or 8, however fast the OEM
moves.  The second factor is the timing of Xenix 3.0, since we want
to combine some elements of Berkeley 4.1 in the release.  I expect
it will be third quarter before XENIX 3.0 is available.  Thus, 68000s
will be available sooner, but with XENIX 2.3.

As for the TRS 68000 machine, it has no memory management in its
current form.  We probably won't do anything with it until a version
with an MMU is available.

I dont understand your question about 'paging restrictions'.  If you
mean virtual memory, we will not support that until we have a real
virtual chip, either the 68010 or the 16032.  One of our OEMs,
MicroDaSys, is porting their own kernal to a 68000, and are doing
a dual-cpu virtual implimentation, where a second CPU wakes up on
a page fault, brings in the page, and lets the second one resume
from its amazingly long memory-access wait state.  I think that this
is a highly specialized system, as it does not provide general
virtual memory performance.

If you mean the fearsome restrictions in the 68000's MMU archetecture
(the Motorolla MMU chip), we've lost a lot of sleep over that design,
and we believe that we have some approaches and techniques to handle
the problem.

As for cost, the price is set by the OEM.  Since our price to them
is quite low, I would expect the OEMs to offer a very attractive price.
Many of them plan on a nice low price for the basic system, with
additional charges for some items which they hold back, such as
(for example) yacc/lex, screen editor, etc.  Our price to the
OEM is very close to the Bell price to us, which was published by
Bell at the last Usenix.  For a 15 user system, its around $300 or
so, roughly.

                                        B.J.
                                        Herbison-BJ@Yale
                                        decvax!yale-comix!herbison-bj
----------