40 years of Unix (BBC) [LWN.net]

40 years of Unix (BBC)

Posted Aug 20, 2009 23:35 UTC (Thu) by jordanb (guest, #45668)
Parent article: 40 years of Unix (BBC)

Multics didn't "fail miserably." I wish that canard would die. Multics *was* very ambitious and *was* delivered quite late. But it was a very powerful, well-engineered piece of software that was well-liked by those who used it -- one with important features that are still missing on Unix systems 40 years later.

The problem for Multics, in the long term, was that it was the operating system for GE Mainframes. GE never cracked the stranglehold IBM had on the mainframe market, and of course mainframes were on their way to becoming dinosaurs anyway.

The coup that Unix pulled off was ending up on DEC machines right as the PDP-11 was taking over the world. That, combined with the completely unplanned and unique arrangement developed with universities due to AT&T's consent decree with the Justice Department, combined with a special interest taken by a group of grad students in Berkeley, resulted in a series of fortunate events for the system, which put it in its trajectory towards ending up the most successful OS in computer history.

to post comments

40 years of Unix (BBC)

Posted Aug 21, 2009 3:51 UTC (Fri) by pabs (subscriber, #43278) [Link] (6 responses)

Which features are missing from Unix that Multics had?

40 years of Unix (BBC)

Posted Aug 21, 2009 4:25 UTC (Fri) by jordanb (guest, #45668) [Link] (5 responses)

Off the top of my head, single level storage, and RPC through dynamic linking. They also had things that are just now getting added to modern Unicies in a very incomplete fashion, like MAC and ACLs,

40 years of Unix (BBC)

Posted Aug 21, 2009 23:13 UTC (Fri) by Wol (subscriber, #4433) [Link] (4 responses)

Don't know to what extent it was hardware, and what software, but Pr1mos (a multics derrivative too) had fast ring-level switching. Essentially, a kernel call with context switch and ring jump was just as fast as a user-level subroutine call. But it was much more secure (look at the recent stuff over jumping to a null pointer - that entire class of problems just can't happen.

Cheers,
Wol

40 years of Unix (BBC)

Posted Aug 25, 2009 2:59 UTC (Tue) by pabs (subscriber, #43278) [Link] (3 responses)

Is anyone working on any of the above features for Linux?

40 years of Unix (BBC)

Posted Aug 25, 2009 21:51 UTC (Tue) by jordanb (guest, #45668) [Link] (2 responses)

Well, MAC and ACLs are getting implemented, slowly, with things like SELinux.

Single-level storage would require fundamental changes in Unix such that it wouldn't be Unix anymore, although this feature does live on in IBM mainframe products (the 'i' and 'z' series, I believe).

Memory-mapped IPC would be possible, I'd think, if processes were able to mmap eachother's executable space and have a linker that could do the binding to the exposed functions on the fly. I don't see why that couldn't be done in a backwards-compatible manner. I also don't know of anyone trying to do it.

40 years of Unix (BBC)

Posted Aug 27, 2009 22:47 UTC (Thu) by giraffedata (guest, #1954) [Link] (1 responses)

Single-level storage would require fundamental changes in Unix such that it wouldn't be Unix anymore, although this feature does live on in IBM mainframe products (the 'i' and 'z' series, I believe).

System I (called Eserver I series from 2001 to 2005, and a bunch of other things before that) does, but the System Z operating systems are about as far as you can get from it.

"mainframe" isn't a very meaningful word these days, but to most people it simply means IBM System Z. Nobody uses the word for System I.

40 years of Unix (BBC)

Posted Aug 28, 2009 2:59 UTC (Fri) by jordanb (guest, #45668) [Link]

Thanks for correction.

I said that because I had read an article by an IBM shill about SLS in IBM products. They *were* speaking specifically about the I-series but kept making statements about how "IBM mainframe users just expect this sort of feature" so I assumed it was more widespread. :<

40 years of Unix (BBC)

Posted Aug 21, 2009 6:20 UTC (Fri) by drag (guest, #31333) [Link] (1 responses)

A operating system nobody can really use that is developed using a machine-specific language for hardware that nobody owns... It would be scary to see a OS that you considered a failure!

> The coup that Unix pulled off was ending up on DEC machines right as the PDP-11 was taking over the world. That, combined with the completely unplanned and unique arrangement developed with universities due to AT&T's consent decree with the Justice Department, combined with a special interest taken by a group of grad students in Berkeley, resulted in a series of fortunate events for the system, which put it in its trajectory towards ending up the most successful OS in computer history.

I always thought it was due to:

* Being relatively simple, so it could run on 'lesser' machines and be used and understood by mortals.
* The early rewrite to C so that it survived it's own original development environment and was used on many different sorts of hardware.
* Having the source code available to anybody that cared.

And later on:

* TCP/IP networking.

I don't think that it was a accident of history. I think that being open and being written in a portable language was critical to it's success.

(and later on the splintering, expense, elitism, and proprietary-ness of commercial Unix later on is what openned the world up for the Wintel take over of everything.)

What other OSes from 30-40 years ago had similar characteristics?

40 years of Unix (BBC)

Posted Aug 21, 2009 14:38 UTC (Fri) by jordanb (guest, #45668) [Link]

> * Being relatively simple, so it could run on 'lesser' machines and be used and understood by mortals.

Yeah, Multics could never run on a PDP-11. So when minicomputers started appearing in every laboratory and corporate department, Unix came with them. Meanwhile Multics was stuck in the Computing Center in multi-million dollar room-sized mainframes. That's my point about Unix "riding the minicomputer wave". :)

> * The early rewrite to C so that it survived it's own original development environment and was used on many different sorts of hardware.

I think that's a mixed bag. On the one hand, Multics was also written in a high level language. On the other hand, PL/1 was a much more powerful language and that made it much more difficult to write a compiler on a new architecture. So one could say that C hit the right abstraction level to maximize portability: It was abstract enough to allow Unix to be largely system-independent while not having itself so many features as to make it difficult to write new C compilers.

There's another part of this though. Multics was a system written at considerable expense by GE as the OS for their mainframes. They controlled the copyright and the source and had no *interest* in seeing it ported to other systems. Just as IBM had no interest in seeing OS/360 run on anything but IBM mainframes.

Unix, on the other hand, was a bit of an orphan. AT&T had no stake in DEC equipment and DEC only grudgingly tolerated it on their machines. Nobody had an interest in keeping it a single-machine OS. And because of AT&T's consent decree, lots of Universities had the code and it was natural for them to port it to their other minicomputers.

> * Having the source code available to anybody that cared.

I agree that this was one of the most important aspects of the success of Unix. But it was the result of the unique sales arrangement AT&T made due to their consent decree with the Justice Department.

As soon as they could they rolled back that back and started trying to sell Unix as a "commercial product," even going so far as to sue the Regents of the University of California to try to kill BSD.

> * TCP/IP networking.

Once again, I think Unix ended up on the right machine at the right time with the right set of social structures around it. Because of the minicomputer revolution, most ARPA participants had easy access to machines that could run Unix. It was natural for them to focus on those machines when building TCP/IP, so Unix naturally ended up with robust internetworking facilities.