Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!agate!helios.ee.lbl.gov!nosc!spectra!pace
From: pace@spectra.COM (William B. Pace)
Newsgroups: comp.os.misc
Subject: Re: Unix bigotry
Summary: Not just ACL's but lightweight processes/threads as well!
Keywords: AOS AOS/VS ACL threads tasks processes
Message-ID: <117@spectra.COM>
Date: 4 Feb 89 20:50:21 GMT
References: <1135@raspail.UUCP> <476@cvbnet2.UUCP> <653@unocss.UUCP>
Organization: Spectragraphics, Corp., San Diego, CA
Lines: 73

In article <653@unocss.UUCP>, mlewis@unocss.UUCP (Marcus S. Lewis) writes:
> In article <476@cvbnet2.UUCP>, aperez@cvbnet2.UUCP (Arturo Perez Ext.) writes:
> [much deleted about acls and such]
> > About the only operating system which had all these features together
> > was multics but it is going the way of the dinosaurs :-(
> 
> I have worked with Data General computers for the last 5 years, and I take
> exception to the statement that only multics has had ACL's and real protections
> etc.  AOS/VS has been around since 1979 with the advent of the MV series of
> processors, and is a natural (virtual) extension of AOS, which is substantially
> older.      [ ...stuff deleted ]

AOS was first introduced circa 1975 along with the Eclipse line of computers.
The Eclipse itself was a natural (non-virtual) extension of the Nova series
and you could see that AOS inherited much of it's philosophy from the RDOS/DOS
operating systems that went with the old Novas.

The AOS ACL's do work quite well.  You can specify pattern matching characters
for classes of usernames.  For instance, if all accounts in a certain project
started (or ended) in the same string (a technique common to many OS's) then
you can match the whole class of names.  Mr. Lewis stated later in his post
that they use the individual's name as the account name.  This makes it harder
since you have to explicitly specify each account but this is a relatively
small detail.  The bad part about ACL's was that you had to re-specify the
ENTIRE access control list whenever you wanted to delete/change/add an entry
(if you used the default CLI).  This detail was an unfortunate oversight
but I guess you can't have everything.  (Do they still do it this way?)

Control Data's NOS also had similar access control (with a PERMIT command)
but with a real nice extra.  Whenever a user accessed a file, the access
was logged by account, number of accesses, date of last access and type
of access.  Even if you PERMITted the file to the world, you could still
get a list of the specific users that accessed the file.  

On another front, I was at USENIX this past week and heard some fascinating
papers on the addition of threads (aka lightweight processes) to Unix.  I'm
fascinated because RDOS had this way back in 1968!  AOS and AOS/VS both
continued the feature which was one of the most FUN parts of these opsystems.
The same features/problems that they were describing at the conference were
settled years ago by DG, or at least were well understood.  I guess none of
the crew that implemented the DG stuff are listening in on the Unix world.
It would be nice if they were to get involved in the process of defining
threads.

AOS(/VS) made a distinction between processes (same as Unix processes) and
tasks (same as threads) and I have always liked their definitions.  Most of
the Unix world equates a process to a task since Unix supports only one task
per process.  There is a rich set of inter-task control and communications
primitives, most of which could be handled within the process context without
a switch to the kernal so they are very efficient.  The problems that needed
to be solved were the same.  Need to do a non-blocking read?  Start a task.

> Non-standard disclaimer:  I don't work for DG, but I'm available....
> Marc Lewis

I DID work for DG about five years ago (when they were hiring all of the
ex-IBM types into upper management and were predicting $1B/year sales the
next year - yea, right) but don't think that biases me: I've done extensive
work in NOS, VMS, MVS, VM/CMS, many others that I've since forgotten, and of
course Unix.  Of all of these, AOS and AOS/VS are my hands-down favorites.
If Unix were AOS, I'd be a happy man.  Unfortunately, there is a limited job
market for practioners of the AOS(/VS) art so I had to make the difficult
decision to leave it.  Sigh.

Those who think that Unix is a 'clean' OS have never looked at AOS code.
THAT's a clean system.  Unix, bless it's tin soul, is NOT a clean OS, despite
what Unix boosters want to say.  Don't get me wrong, next to AOS, Unix is my
favorite - but that doesn't blind me to it's faults.  (Flames by mail please).

Bill Pace
pace@spectra.com

Disclaimer:  I don't disclaim a damn thing!