Was there ever a PDP-10 unix project that completed?

[RSX11M+]

My days with the PDP-10 go back to my time at Stevens Institute.

In that epoch, Timesharing and Batch were the most common system methodologies. There was talk of some outgrowth of research at Bell Labs, called unix, but it was mostly ignored in the beginning.

Digital [DEC] was "pushing" their PDP-11 line as the "coming thing" but being a paltry 16 bits, it wasn't targeted at large systems. [Don't tell that to the 11/70 group]

Fast forward a few years, and 36 bit systems seemed to be orphaned. The PDP-11 line itself was even seen as "diminishing", yielding to something called VAX. The preeminent 11 operating system was RSX or unix, depending on which camp you came from - with RT-11 coming in third. [RSX-11M+ was in it's infancy]

There was a definite schism between these opposing camps [tribes might be more appropriate]. "Realtimers" [as we knew ourselves] detested VAX. We constantly heard that an 11/780 would be able to run PDP-11 applications in "compatibility mode", but it never did. We were told it would be faster than the fastest 11, but the 11/70 always tested faster... much faster.

Anyway, curiously, we Realtimers looked up, not to VMS [VAX] but to the 10s and their world. Big systems, bigger worlds - REAL machines. VAX would never measure up. Not even when they tried to shove one up our ___ with something called "VAXelan".

The VAX camp[ers] in turn detested the PDP-10 groups. [These included PDP-6, DECsystem 10 and 20 units] They continuously boasted of their bright future, despite evidence to the contrary. To them, PDP-11s were no longer even a "necessary evil", and considered passé. [In the VAX world, Realtime meant "This week"]


Given these primitive conditions and hierarchy, few would even consider committing the sacrilege of running such an antithetical OS as unix on a PDP-10. However, some were thinking just that. I've heard of several implimentations of 'C' for 10s, but only rumors of ports of unix.

Now we come to the point of my inquiry...

Is anyone aware of a serious attempt to port unix for a 36 bit system? :confused1:
​

I'd really like to know, and be curious to hear how it was tailored for that machine, even if it were ultimately abandoned. Tell me a story, spread another rumor, speculate even... just try not to fib too much.



Yeah... I'm only a couple years behind... been spending a lot of time in "memory lane" lately. :lookroun:

 

[krebizfan]

According to set of posts in alt.folklore.computers in 2007, there was a 36-bit Unix written for the Univac 1100. But that seems to be the only 36-bit port that anyone knows of.

 

[RSX11M+]

I figured of anyone, COMPUSERVe would have. [or perhaps hold-outs at Xerox PARC]

From the Wikipedia:

As of January 2007, CompuServe continues to operate a small number of PDP-10 architecture machines to perform some billing and routing functions. The main power supplies used in the KL-series machines were so inefficient that CompuServe engineers designed a replacement power supply that consumed about half the energy. CompuServe offered to license the design for its KL power supply to DEC for free if DEC would promise that any new KL purchased by CompuServe would have the more efficient power supply installed. DEC declined the offer.

They certainly would have had long enough to consider and complete it.

Anyway, I didn't restrict myself to DEC 36-bit systems, did I?

UNIVAC 1100... never would have thought.

 

[tradde]

I figured of anyone, COMPUSERVe would have. [or perhaps hold-outs at Xerox PARC]

From the Wikipedia:



They certainly would have had long enough to consider and complete it.

Anyway, I didn't restrict myself to DEC 36-bit systems, did I?

UNIVAC 1100... never would have thought.

I never heard of it when I worked at Unisys from 1981 to 2000 (well, I worked for Sperry from 1981 to 1986 when the merger happened).

 

[Chuck(G)]

Doesn't C pretty much demand twos complement arithmetic? ISTR that the 1100 series was ones' complement. Your choice of 6 or 9 bit characters, too.

I mention this because at one point I recall asking some of the Unix gurus if Unix would work on a ones' complement machine. The off-the-cuff opinion was "probably not without a lot of trouble".

So, my next question would be "Did anyone ever port Unix to a variable word length machine?" Or a decimal machine?

 

[krebizfan]

I remembered some things that lead to a successful search on the UNIVAC with Unix. It is a description of what had to be done to port Unix. The amount of debugging needed and working around assumptions in Unix makes me happy that my programming career avoided any similar project.

http://cm.bell-labs.com/cm/cs/who/dmr/otherports/newp.html look at section IV

 

[Chuck(G)]

I see that the 1100 port was a cheat of sorts, in that Unix ran as a task under the resident OS, EXEC. I wonder how functional the result was--and how robust.

Are there any really obscure ports, such as IBM 7080 or Univac LARC?

 

[RSX11M+]

I was researching your posts and something odd happened... I'm wondering if it's just me.

I'd loaded Dennis Ritchie's home page, and was following links [several successfully] when they started timing out.

Looking into it, I suddenly find that the home page domain cm.bell-labs.com or bell-labs.com is no longer registered in DNS...??!?

Is anyone else seeing this?

It's pretty much inconceivable, so I'm thinking it must be a DNS exploit or something.

If you get a response from either domain, please post the numeric IP address here. - Thanks


UPDATE:
I just switched to openDNS and got the same basic result.

 

[Al Kossow]

I never heard of it when I worked at Unisys from 1981 to 2000 (well, I worked for Sperry from 1981 to 1986 when the merger happened).

It was done at Bell Labs, Indian Hill and ran on top of Exec-8
Univ of Wisconsin - Milwaukee experimented with it in the early 80s, which is
where I saw it. I don't know if it ever went into production there.

So, as far as I know, there were no 36 bit ports onto bare hardware.

 

[nozomi]

I was researching your posts and something odd happened... I'm wondering if it's just me.

I'd loaded Dennis Ritchie's home page, and was following links [several successfully] when they started timing out.

Looking into it, I suddenly find that the home page domain cm.bell-labs.com or bell-labs.com is no longer registered in DNS...??!?

Is anyone else seeing this?

It's pretty much inconceivable, so I'm thinking it must be a DNS exploit or something.

If you get a response from either domain, please post the numeric IP address here. - Thanks


UPDATE:
I just switched to openDNS and got the same basic result.

host cm.bell-labs.com
cm.bell-labs.com CNAME plan9.bell-labs.com
plan9.bell-labs.com A 204.178.31.32

 

[RSX11M+]

host cm.bell-labs.com
cm.bell-labs.com CNAME plan9.bell-labs.com
plan9.bell-labs.com A 204.178.31.32

Thank you... I was beginning to think that post was invisible - LOL.

Well, I see it's at least that far again [DNS response], but now his homepage errs...

[h=1]Object not found[/h] The object /who/dmr/ does not exist on this server. errstr: '/usr/web/who' Hangup
uri host:
header host: cm.bell-labs.com
actual host: plan9.bell-labs.com

I guess the problem is real. Just a weird coincidence - but that's my life.

 

[RSX11M+]

I remembered some things that lead to a successful search on the UNIVAC with Unix. It is a description of what had to be done to port Unix. The amount of debugging needed and working around assumptions in Unix makes me happy that my programming career avoided any similar project.

http://cm.bell-labs.com/cm/cs/who/dmr/otherports/newp.html look at section IV

That's a wonderful link - just the kind of insightful thought provoking discussion I was looking for - thank you.

But no talk of it for a DEC 10 eh?

Following Dennis Ritchie's homepage links, [until the DNS problem happened!] I was also able to download an early PCC [Portable C Compiler] source. It's something I've always wanted to see, and goes a long way to expose the fundamental architectural issues of such a port.

Thank you all... great thread so far.

 

[ahm]

I figured of anyone, COMPUSERVe would have.

But you need to ask yourself - why would they?

They'd already built their stuff to run on TOPS.
So unless it would have made them more money, there's no motivation.
Developing new software or porting old software costs money.

 

[Chuck(G)]

So what's the smallest (in terms of computing power and memory) processor to which Unix has been ported?

 

[Al Kossow]

So what's the smallest (in terms of computing power and memory) processor to which Unix has been ported?

Morrow ported V6 to the Z80 with a special MMU (MPZ80 card)

 

[Chuck(G)]

Nothing smaller? Not even on the "what's so special about the Z80?" 6502? I guess in theory, you could port it to any Turing-complete processor, given enough storage.

 

[RSX11M+]

But you need to ask yourself - why would they?

They'd already built their stuff to run on TOPS.
So unless it would have made them more money, there's no motivation.
Developing new software or porting old software costs money.

"Why would they?"

Well, lets see... I thought the answer was self-evident, but this is the way I was looking at it...

1) Compuserve was a computing services company. Meaning - they sold users [businesses] time on their systems to run applications.

2) Compuserve was in the business of making money on it's programming talent, and hosing the applications on their systems.

3) Over time, Compuserve stuck with the PDP-10 longer than anyone else. They continued to develop their 10s into a nation-wide ARPANET network. They designed their own hardware when it wasn't available and improved their machines out of their own talents, making huge sums for H&R Block as they did.

4) During the time that Compuserve was thriving, the unix operating system too, was evolving into a viable commercial platform, particularly after POSIX. Windows and MAC machines had not taken over the planet. The only thing that had not gotten going yet was an opensource unix. This would eventually happen, but not in time to inject itself into Compuserve's DNA until the late 80s. In the 90s, even MACs transitioned to a unix based OS.

5) The UNIVAC 1100 solution shows at least one way in which unix services could coexist with, and run in conjunction with a host system such as TOPS or TENEX. This would facilitate offering unix based application services with ready made ties to the 10s other networking and systems support.
​

Along this timeline, several viable 'C' compilers for the PDP-10 were produced. Compuserve's resources were sufficient that they could purchase any one of them had they wanted to pursue this course.

The only thing that was apparently missing was the wisdom to make it happen. [since it didn't?]


I really didn't know what I'd hear when I posed this question. I had an expectation that someone, somewhere might know of an effort in this direction if it ever took place. As I said, given all that about Compuserve... well, you know.

BTW - There was a rumor that went around in the early 80's that still boggles the mind. The story went that AT&T, offered the unix operating system to Digital Equipment Corporation, FREE - on the sole proviso that DEC support it. Some genius at DEC decided it was a system in decline, and turned them down.

 

[Al Kossow]

Nothing smaller?

I would think an 8 bit processor, 64K address space and a 16 bit stack pointer, along with a small kernel, like Version 6 is
about the bottom limit. You'd have a tough time even getting Version 7 applications running on something that small.

There was a version of Xenix that ran on the XT also, but at least that had segmentation so you could get above 64k.

 

[RSX11M+]

On the "small" side, there's a Bell System Technical Journal article from 1977 about a single user unix on LSI-11 [PDP-11/02?] called LSX.

In it, they talk about a system kernel occupying 8KW of memory and able to run reasonable apps in 20KW.

Of course, this would still be considered a 16-bit CPU. [so was the XT] Well, I would, since we always sized the CPU based on the ALU, not the databus width. If we're only considering the "bus size" to qualify it as an 8-bit machine, then the DCT-11 running in 8-bit mode would also qualify, while retaining code compatibility with LSX.

SCO XENIX System 5 ran on an 8088/8086 PC (1983).

One could argue that some large microprocessor CPUs are implemented on bit-slice hardware running microcode, so in a sense the "smallness" of the CPU is "blurry".

I do have a vague recollection of CPU with a "bit serial bus" that ran a unix clone OS for an aerospace app. Wish I could recall enough to look for it. Used a whole collection of I/O and memory with the same bus. [kind of like early XILINX stuff] Really interesting concept with ultra high pin-density hardware. I'm thinking pre 1986? With such a limited scope, I'm not sure you'd call it enough of a system to qualify as full unix.

Still, Z80 is genuinely 8 bit / 64KB. I don't consider an MMU as a requirement to call it "real" unix, do you?

 

[Chuck(G)]

Well, with a disk, you can pull all sorts of tricks. Remember JRT Pascal? Virtual memory on an 8080, so anything's possible.

I consider 6502, 6800 and less-used CPUs such as SC/MP to be "smaller" than a Z80. Probably likewise for a PDP-8, even though it was a 12-bit machine--the instruction set and addressability is very limited.