LINUX is obsolete (original) (raw)

Andy Tanenbaum

unread,

Jan 29, 1992, 7:12:50 AM1/29/92

I was in the U.S. for a couple of weeks, so I haven't commented much on
LINUX (not that I would have said much had I been around), but for what
it is worth, I have a couple of comments now.

As most of you know, for me MINIX is a hobby, something that I do in the
evening when I get bored writing books and there are no major wars,
revolutions, or senate hearings being televised live on CNN. My real
job is a professor and researcher in the area of operating systems.

As a result of my occupation, I think I know a bit about where operating
are going in the next decade or so. Two aspects stand out:

1. MICROKERNEL VS MONOLITHIC SYSTEM
Most older operating systems are monolithic, that is, the whole operating
system is a single a.out file that runs in 'kernel mode.' This binary
contains the process management, memory management, file system and the
rest. Examples of such systems are UNIX, MS-DOS, VMS, MVS, OS/360,
MULTICS, and many more.

The alternative is a microkernel-based system, in which most of the OS
runs as separate processes, mostly outside the kernel. They communicate
by message passing. The kernel's job is to handle the message passing,
interrupt handling, low-level process management, and possibly the I/O.
Examples of this design are the RC4000, Amoeba, Chorus, Mach, and the
not-yet-released Windows/NT.

While I could go into a long story here about the relative merits of the
two designs, suffice it to say that among the people who actually design
operating systems, the debate is essentially over. Microkernels have won.
The only real argument for monolithic systems was performance, and there
is now enough evidence showing that microkernel systems can be just as
fast as monolithic systems (e.g., Rick Rashid has published papers comparing
Mach 3.0 to monolithic systems) that it is now all over but the shoutin`.

MINIX is a microkernel-based system. The file system and memory management
are separate processes, running outside the kernel. The I/O drivers are
also separate processes (in the kernel, but only because the brain-dead
nature of the Intel CPUs makes that difficult to do otherwise). LINUX is
a monolithic style system. This is a giant step back into the 1970s.
That is like taking an existing, working C program and rewriting it in
BASIC. To me, writing a monolithic system in 1991 is a truly poor idea.

2. PORTABILITY
Once upon a time there was the 4004 CPU. When it grew up it became an
8008. Then it underwent plastic surgery and became the 8080. It begat
the 8086, which begat the 8088, which begat the 80286, which begat the
80386, which begat the 80486, and so on unto the N-th generation. In
the meantime, RISC chips happened, and some of them are running at over
100 MIPS. Speeds of 200 MIPS and more are likely in the coming years.
These things are not going to suddenly vanish. What is going to happen
is that they will gradually take over from the 80x86 line. They will
run old MS-DOS programs by interpreting the 80386 in software. (I even
wrote my own IBM PC simulator in C, which you can get by FTP from
ftp.cs.vu.nl = 192.31.231.42 in dir minix/simulator.) I think it is a
gross error to design an OS for any specific architecture, since that is
not going to be around all that long.

MINIX was designed to be reasonably portable, and has been ported from the
Intel line to the 680x0 (Atari, Amiga, Macintosh), SPARC, and NS32016.
LINUX is tied fairly closely to the 80x86. Not the way to go.

Don`t get me wrong, I am not unhappy with LINUX. It will get all the people
who want to turn MINIX in BSD UNIX off my back. But in all honesty, I would
suggest that people who want a **MODERN** "free" OS look around for a
microkernel-based, portable OS, like maybe GNU or something like that.

Andy Tanenbaum (a...@cs.vu.nl)

P.S. Just as a random aside, Amoeba has a UNIX emulator (running in user
space), but it is far from complete. If there are any people who would
like to work on that, please let me know. To run Amoeba you need a few 386s,
one of which needs 16M, and all of which need the WD Ethernet card.

David Megginson

unread,

Jan 29, 1992, 9:12:12 AM1/29/92

I would like to at least look at LINUX, but I cannot, since I run
a 68000-based machine. In any case, it is nice having the kernel
independent, since patches like the multi-threaded FS patch don't
have to exist in a different version for each CPU.

I second everything AST said, except that I would like to see
the kernel _more_ independent from everything else. Why does the
Intel architecture _not_ allow drivers to be independent programs?

I also don't like the fact that the kernel, mm and fs share the
same configuration files. Since they _are_ independent, they should
have more of a sense of independence.

David

#################################################################
David Megginson meg...@epas.utoronto.ca
Centre for Medieval Studies da...@doe.utoronto.ca
University of Toronto 39 Queen's Park Cr. E.
#################################################################

Andy Tanenbaum

unread,

Jan 29, 1992, 1:03:01 PM1/29/92

The drivers have to read and write the device registers in I/O space, and
this cannot be done in user mode on the 286 and 386. If it were possible
to do I/O in a protected way in user space, all the I/O tasks could have
been user programs, like FS and MM.

Andy Tanenbaum (a...@cs.vu.nl)

Kevin Brown

unread,

Jan 29, 1992, 8:36:43 PM1/29/92

Of course, there are some things that are best left to the kernel, be it
micro or monolithic. Like things that require playing with the process'
stack, e.g. signal handling. Like memory allocation. Things like that.

The microkernel design is probably a win, all in all, over a monolithic
design, but it depends on what you put in the kernel and what you leave
out.

> MINIX is a microkernel-based system. The file system and memory management
> are separate processes, running outside the kernel. The I/O drivers are
> also separate processes (in the kernel, but only because the brain-dead
> nature of the Intel CPUs makes that difficult to do otherwise).

Minix is a microkernel design, of sorts. The problem is that it gives special
priveleges to mm and fs, when there shouldn't be any (at least for fs). It
also fails to integrate most of the functionality of mm in the kernel itself,
and this makes things like signal handling and memory allocation *really*
ugly. If you did these things in the kernel itself, then signal handling
would be as simple as setting a virtual interrupt vector and causing the
signalled process to receive that interrupt (with the complication that
system calls might have to be terminated. Which means that a message would
have to be sent to every process that is servicing the process' system call,
if any. It's considerations like these that make the monolithic kernel
design appealing).

The *entire* system call interface in Minix needs to be rethought. As it
stands right now, the file system is not just a file system, it's also a
system-call server. That functionality needs to be separated out in order
to facilitate a multiple file system architecture. Message passing is
probably the right way to go about making the call and waiting for it, but
the message should go to a system call server, not the file system itself.

In order to handle all the special caveats of the Unix API, you end up writing
a monolithic "kernel" even if you're using a microkernel base. You end up
with something called a "server", and an example is the BSD server that runs
under Mach.

And, in any case, the message-passing in Minix needs to be completely redone.
As it is, it's a kludge. I've been giving this some thought, but I haven't
had time to do anything with what I've thought of so far. Suffice it to say
that the proper way to do message-passing is probably with message ports
(both public and private), with the various visible parts of the operating
system having public message ports. Chances are, that ends up being the
system call server only, though this will, of course, depend on the goals
of the design.

> LINUX is
> a monolithic style system. This is a giant step back into the 1970s.
> That is like taking an existing, working C program and rewriting it in
> BASIC. To me, writing a monolithic system in 1991 is a truly poor idea.

Depends on the design criteria, as you should know. If your goal is to
design a Unix workalike that is relatively simple and relatively small,
then a monolithic design is probably the right approach for the job, because
unless you're designing for really backwards hardware, the problems of
things like interrupted system calls, memory allocation within the kernel
(so you don't have to statically allocate *everything* in your OS), signal
handling, etc. all go away (or are at least minimized) if you use a
monolithic design. If you want the ability to bring up and take down
file systems, add and remove device drivers, etc., all at runtime, then
a microkernel approach is the right solution.

Frankly, I happen to like the idea of removable device drivers and such,
so I tend to favor the microkernel approach as a general rule.

>2. PORTABILITY
> Once upon a time there was the 4004 CPU. When it grew up it became an
> 8008. Then it underwent plastic surgery and became the 8080. It begat
> the 8086, which begat the 8088, which begat the 80286, which begat the
> 80386, which begat the 80486, and so on unto the N-th generation. In
> the meantime, RISC chips happened, and some of them are running at over
> 100 MIPS. Speeds of 200 MIPS and more are likely in the coming years.
> These things are not going to suddenly vanish. What is going to happen
> is that they will gradually take over from the 80x86 line. They will
> run old MS-DOS programs by interpreting the 80386 in software. (I even
> wrote my own IBM PC simulator in C, which you can get by FTP from
> ftp.cs.vu.nl = 192.31.231.42 in dir minix/simulator.) I think it is a
> gross error to design an OS for any specific architecture, since that is
> not going to be around all that long.

Again, look at the design criteria. If portability isn't an issue, then
why worry about it? While LINUX suffers from lack of portability, portability
was obviously never much of a consideration for its author, who explicitly
stated that it was written as an exercise in learning about the 386
architecture.

And, in any case, while MINIX is portable in the sense that most of the code
can be ported to other platforms, it *still* suffers from the limitations of
the original target machine that drove the walk down the design decision tree.
The message passing is a kludge because the 8088 is slow. The kernel doesn't
do memory allocation (thus not allowing FS and the drivers to get away with
using a malloc library or some such, and thus causing everyone to have to
statically allocate everything), probably due to some other limitation of
the 8088. The very idea of using "clicks" is obviously the result of the
segmented architecture of the 8088. The file system size is too limited
(theoretically fixed in 1.6, but now you have *two* file system formats to
contend with. If having the file system as a separate process is such a
big win, then why don't we have two file system servers, eh? Why simply
extend the existing Minix file system instead of implementing BSD's FFS
or some other high-performance file system? It's not that I'm greedy
or anything... :-).

> MINIX was designed to be reasonably portable, and has been ported from the
> Intel line to the 680x0 (Atari, Amiga, Macintosh), SPARC, and NS32016.
> LINUX is tied fairly closely to the 80x86. Not the way to go.

All in all, I tend to agree.

>Don`t get me wrong, I am not unhappy with LINUX. It will get all the people
>who want to turn MINIX in BSD UNIX off my back. But in all honesty, I would
>suggest that people who want a **MODERN** "free" OS look around for a
>microkernel-based, portable OS, like maybe GNU or something like that.

Yeah, right. Point me someplace where I can get a free "modern" OS and I'll
gladly investigate. But the GNU OS is currently vaporware, and as far as I'm
concerned it will be for a LOOOOONG time to come.

Any other players? BSD 4.4 is a monolithic architecture, so by your
definition it's out. Mach is free, but the BSD server isn't (AT&T code,
you know), and in any case, isn't the BSD server something you'd consider
to be a monolithic design???

Really. Why do you think LINUX is as popular as it is? The answer is
simple, of course: because it's the *only* free Unix workalike OS in
existence. BSD doesn't qualify (yet). Minix doesn't qualify. XINU
isn't even in the running. GNU's OS is vaporware, and probably will
be for a long time, so *by definition* it's not in the running. Any
other players? I haven't heard of any...

>Andy Tanenbaum (a...@cs.vu.nl)

Minix is an excellent piece of work. A good starting point for anyone who
wants to learn about operating systems. But it needs rewriting to make it
truly elegant and functional. As it is, there are too many kludges and
hacks (e.g., the message passing).

Kevin Brown

Linus Benedict Torvalds

unread,

Jan 29, 1992, 6:14:26 PM1/29/92

Well, with a subject like this, I'm afraid I'll have to reply.
Apologies to minix-users who have heard enough about linux anyway. I'd
like to be able to just "ignore the bait", but ... Time for some
serious flamefesting!

>I was in the U.S. for a couple of weeks, so I haven't commented much on
>LINUX (not that I would have said much had I been around), but for what
>it is worth, I have a couple of comments now.
>
>As most of you know, for me MINIX is a hobby, something that I do in the
>evening when I get bored writing books and there are no major wars,
>revolutions, or senate hearings being televised live on CNN. My real
>job is a professor and researcher in the area of operating systems.

You use this as an excuse for the limitations of minix? Sorry, but you
loose: I've got more excuses than you have, and linux still beats the
pants of minix in almost all areas. Not to mention the fact that most
of the good code for PC minix seems to have been written by Bruce Evans.

Re 1: you doing minix as a hobby - look at who makes money off minix,
and who gives linux out for free. Then talk about hobbies. Make minix
freely available, and one of my biggest gripes with it will disappear.
Linux has very much been a hobby (but a serious one: the best type) for
me: I get no money for it, and it's not even part of any of my studies
in the university. I've done it all on my own time, and on my own
machine.

Re 2: your job is being a professor and researcher: That's one hell of a
good excuse for some of the brain-damages of minix. I can only hope (and
assume) that Amoeba doesn't suck like minix does.

>1. MICROKERNEL VS MONOLITHIC SYSTEM

True, linux is monolithic, and I agree that microkernels are nicer. With
a less argumentative subject, I'd probably have agreed with most of what
you said. From a theoretical (and aesthetical) standpoint linux looses.
If the GNU kernel had been ready last spring, I'd not have bothered to
even start my project: the fact is that it wasn't and still isn't. Linux
wins heavily on points of being available now.

> MINIX is a microkernel-based system. [deleted, but not so that you
> miss the point ] LINUX is a monolithic style system.

If this was the only criterion for the "goodness" of a kernel, you'd be
right. What you don't mention is that minix doesn't do the micro-kernel
thing very well, and has problems with real multitasking (in the
kernel). If I had made an OS that had problems with a multithreading
filesystem, I wouldn't be so fast to condemn others: in fact, I'd do my
damndest to make others forget about the fiasco.

[ yes, I know there are multithreading hacks for minix, but they are
hacks, and bruce evans tells me there are lots of race conditions ]

>2. PORTABILITY

"Portability is for people who cannot write new programs"
-me, right now (with tongue in cheek)

The fact is that linux is more portable than minix. What? I hear you
say. It's true - but not in the sense that ast means: I made linux as
conformant to standards as I knew how (without having any POSIX standard
in front of me). Porting things to linux is generally /much/ easier
than porting them to minix.

I agree that portability is a good thing: but only where it actually has
some meaning. There is no idea in trying to make an operating system
overly portable: adhering to a portable API is good enough. The very
/idea/ of an operating system is to use the hardware features, and hide
them behind a layer of high-level calls. That is exactly what linux
does: it just uses a bigger subset of the 386 features than other
kernels seem to do. Of course this makes the kernel proper unportable,
but it also makes for a /much/ simpler design. An acceptable trade-off,
and one that made linux possible in the first place.

I also agree that linux takes the non-portability to an extreme: I got
my 386 last January, and linux was partly a project to teach me about
it. Many things should have been done more portably if it would have
been a real project. I'm not making overly many excuses about it
though: it was a design decision, and last april when I started the
thing, I didn't think anybody would actually want to use it. I'm happy
to report I was wrong, and as my source is freely available, anybody is
free to try to port it, even though it won't be easy.

Linus

PS. I apologise for sometimes sounding too harsh: minix is nice enough
if you have nothing else. Amoeba might be nice if you have 5-10 spare
386's lying around, but I certainly don't. I don't usually get into
flames, but I'm touchy when it comes to linux :)

Louie

unread,

Jan 29, 1992, 9:55:22 PM1/29/92

>But in all honesty, I would
>suggest that people who want a **MODERN** "free" OS look around for a
>microkernel-based, portable OS, like maybe GNU or something like that.

There are really no other alternatives other than Linux for people like
me who want a "free" OS. Considering that the majority of people who
would use a "free" OS use the 386, portability is really not all that
big of a concern. If I had a Sparc I would use Solaris.

As it stands, I installed Linux with gcc, emacs 18.57, kermit and all of the
GNU utilities without any trouble at all. No need to apply patches. I
just followed the installation instructions. I can't get an OS like
this *anywhere* for the price to do my Computer Science homework. And
it seems like network support and then X-Windows will be ported to Linux
well before Minix. This is something that would be really useful. In my
opinion, portability of standard Unix software is important also.

I know that the design using a monolithic system is not as good as the
microkernel. But for the short term future (And I know I won't/can't
be uprading from my 386), Linux suits me perfectly.

Philip Wu
p...@unixg.ubc.ca

Jim Burns

unread,

Jan 29, 1992, 10:39:48 PM1/29/92

> The drivers have to read and write the device registers in I/O space, and
> this cannot be done in user mode on the 286 and 386. If it were possible
> to do I/O in a protected way in user space, all the I/O tasks could have
> been user programs, like FS and MM.

The standard way of doing that is to trap on i/o space protection
violations, and emulate the i/o for the user.
--
BURNS,JIM (returned student)
Georgia Institute of Technology, 30178 Georgia Tech Station,
Atlanta Georgia, 30332 | Internet: gt0...@prism.gatech.edu
uucp: ...!{decvax,hplabs,ncar,purdue,rutgers}!gatech!prism!gt0178a

Andy Tanenbaum

unread,

Jan 30, 1992, 8:44:34 AM1/30/92

In article <1992Jan29.2...@klaava.Helsinki.FI> torv...@klaava.Helsinki.FI (Linus Benedict Torvalds) writes:
>You use this [being a professor] as an excuse for the limitations of minix?
The limitations of MINIX relate at least partly to my being a professor:
An explicit design goal was to make it run on cheap hardware so students
could afford it. In particular, for years it ran on a regular 4.77 MHZ PC
with no hard disk. You could do everything here including modify and recompile
the system. Just for the record, as of about 1 year ago, there were two
versions, one for the PC (360K diskettes) and one for the 286/386 (1.2M).
The PC version was outselling the 286/386 version by 2 to 1. I don't have
figures, but my guess is that the fraction of the 60 million existing PCs that
are 386/486 machines as opposed to 8088/286/680x0 etc is small. Among students
it is even smaller. Making software free, but only for folks with enough money
to buy first class hardware is an interesting concept.
Of course 5 years from now that will be different, but 5 years from now
everyone will be running free GNU on their 200 MIPS, 64M SPARCstation-5.

>Re 2: your job is being a professor and researcher: That's one hell of a
>good excuse for some of the brain-damages of minix. I can only hope (and
>assume) that Amoeba doesn't suck like minix does.

Amoeba was not designed to run on an 8088 with no hard disk.

>If this was the only criterion for the "goodness" of a kernel, you'd be
>right. What you don't mention is that minix doesn't do the micro-kernel
>thing very well, and has problems with real multitasking (in the
>kernel). If I had made an OS that had problems with a multithreading
>filesystem, I wouldn't be so fast to condemn others: in fact, I'd do my
>damndest to make others forget about the fiasco.

A multithreaded file system is only a performance hack. When there is only
one job active, the normal case on a small PC, it buys you nothing and adds
complexity to the code. On machines fast enough to support multiple users,
you probably have enough buffer cache to insure a hit cache hit rate, in
which case multithreading also buys you nothing. It is only a win when there
are multiple processes actually doing real disk I/O. Whether it is worth
making the system more complicated for this case is at least debatable.

I still maintain the point that designing a monolithic kernel in 1991 is
a fundamental error. Be thankful you are not my student. You would not
get a high grade for such a design :-)

>The fact is that linux is more portable than minix. What? I hear you
>say. It's true - but not in the sense that ast means: I made linux as
>conformant to standards as I knew how (without having any POSIX standard
>in front of me). Porting things to linux is generally /much/ easier
>than porting them to minix.

MINIX was designed before POSIX, and is now being (slowly) POSIXized as
everyone who follows this newsgroup knows. Everyone agrees that user-level
standards are a good idea. As an aside, I congratulate you for being able
to write a POSIX-conformant system without having the POSIX standard in front
of you. I find it difficult enough after studying the standard at great length.

My point is that writing a new operating system that is closely tied to any
particular piece of hardware, especially a weird one like the Intel line,
is basically wrong. An OS itself should be easily portable to new hardware
platforms. When OS/360 was written in assembler for the IBM 360
25 years ago, they probably could be excused. When MS-DOS was written
specifically for the 8088 ten years ago, this was less than brilliant, as
IBM and Microsoft now only too painfully realize. Writing a new OS only for the
386 in 1991 gets you your second 'F' for this term. But if you do real well
on the final exam, you can still pass the course.

Prof. Andrew S. Tanenbaum (a...@cs.vu.nl)

David Feustel

unread,

Jan 30, 1992, 1:57:28 PM1/30/92

a...@cs.vu.nl (Andy Tanenbaum) writes:

>I still maintain the point that designing a monolithic kernel in 1991 is
>a fundamental error. Be thankful you are not my student. You would not
>get a high grade for such a design :-)

That's ok. Einstein got lousy grades in math and physics.
--
David Feustel N9MYI, 1930 Curdes Ave, Fort Wayne, IN 46805. (219)482-9631
feu...@netcom.com
=== NBC News: GE's Advertising And Public Relations Agency ===

David Megginson

unread,

Jan 30, 1992, 2:58:50 PM1/30/92

In article 1992Jan30.1857...@netcom.COM feu...@netcom.COM (David Feustel) writes:
>a...@cs.vu.nl (Andy Tanenbaum) writes:
>
>
>>I still maintain the point that designing a monolithic kernel in 1991 is
>>a fundamental error. Be thankful you are not my student. You would not
>>get a high grade for such a design :-)
>
>That's ok. Einstein got lousy grades in math and physics.

And Dan Quayle got low grades in political science. I think that there
are more Dan Quayles than Einsteins out there... ;-)

Randy Burns

unread,

Jan 30, 1992, 3:33:07 PM1/30/92

>Of course 5 years from now that will be different, but 5 years from now
>everyone will be running free GNU on their 200 MIPS, 64M SPARCstation-5.

Well, I for one would _love_ to see this happen.

........

>My point is that writing a new operating system that is closely tied to any
>particular piece of hardware, especially a weird one like the Intel line,
>is basically wrong.

First off, the parts of Linux tuned most finely to the 80x86 are the Kernel
and the devices. My own sense is that even if Linux is simply a stopgap
measure to let us all run GNU software, it is still worthwhile to have a
a finely tuned kernel for the most numerous architecture presently in
existance.

> An OS itself should be easily portable to new hardware
>platforms.

Well, the only part of Linux that isn't portable is the kernel and drivers.
Compare to the compilers, utilities, windowing system etc. this is really
a small part of the effort. Since Linux has a large degree of call
compatibility with portable OS's I wouldn't complain. I'm personally
very grateful to have an OS that makes it more likely that some of us will
be able to take advantage of the software that has come out of Berkeley,
FSF, CMU etc. It may well be that in 2-3 years when ultra cheap BSD
variants and Hurd proliferate, that Linux will be obsolete. Still, right
now Linux greatly reduces the cost of using tools like gcc, bison, bash
which are useful in the development of such an OS.

Linus Benedict Torvalds

unread,

Jan 30, 1992, 10:38:16 AM1/30/92

And reply I did, with complete abandon, and no thought for good taste
and netiquette. Apologies to ast, and thanks to John Nall for a friendy
"that's not how it's done"-letter. I over-reacted, and am now composing
a (much less acerbic) personal letter to ast. Hope nobody was turned
away from linux due to it being (a) possibly obsolete (I still think
that's not the case, although some of the criticisms are valid) and (b)
written by a hothead :-)

Linus "my first, and hopefully last flamefest" Torvalds

David Feustel

unread,

Jan 30, 1992, 6:15:05 PM1/30/92

meg...@epas.utoronto.ca (David Megginson) writes:

>In article 1992Jan30.1857...@netcom.COM feu...@netcom.COM (David Feustel) writes:
>>a...@cs.vu.nl (Andy Tanenbaum) writes:
>>
>>
>>>I still maintain the point that designing a monolithic kernel in 1991 is
>>>a fundamental error. Be thankful you are not my student. You would not
>>>get a high grade for such a design :-)
>>
>>That's ok. Einstein got lousy grades in math and physics.

>And Dan Quayle got low grades in political science. I think that there
>are more Dan Quayles than Einsteins out there... ;-)

But the Existence of Linux suggests that we may have more of an
Einstein than a Quail here.

Geoff Collyer

unread,

Jan 30, 1992, 8:13:00 PM1/30/92

Andy Tanenbaum:

>MINIX was designed before POSIX, and is now being (slowly) POSIXized as
>everyone who follows this newsgroup knows.

May I recommend the use of the verb "posixiate" (by analogy with
asphyxiate) instead of "posixize"? Similarly, I prefer "ansitise"
(converse and anagram of "sanitise") to "ansify".
--
Geoff Collyer world.std.com!geoff, uunet.uu.net!geoff

Kevin Brown

unread,

Jan 31, 1992, 2:43:47 AM1/31/92

Sorry, but I just can't resist this thread...:-)

>In article <1992Jan29.2...@klaava.Helsinki.FI> torv...@klaava.Helsinki.FI (Linus Benedict Torvalds) writes:
>>You use this [being a professor] as an excuse for the limitations of minix?
>The limitations of MINIX relate at least partly to my being a professor:
>An explicit design goal was to make it run on cheap hardware so students
>could afford it. In particular, for years it ran on a regular 4.77 MHZ PC
>with no hard disk.

And an explicit design goal of Linux was to take advantage of the special
features of the 386 architecture. So what exactly is your point? Different
design goals get you different designs. You ought to know that.

>You could do everything here including modify and recompile
>the system. Just for the record, as of about 1 year ago, there were two
>versions, one for the PC (360K diskettes) and one for the 286/386 (1.2M).
>The PC version was outselling the 286/386 version by 2 to 1. I don't have
>figures, but my guess is that the fraction of the 60 million existing PCs that
>are 386/486 machines as opposed to 8088/286/680x0 etc is small. Among students
>it is even smaller.

I find it very interesting that you claim here that Minix was designed
primarily for cheap hardware (in particular, the IBM PC/XT with no hard
disk) and yet elsewhere have also mentioned the virtues of being portable
across hardware platforms. Well, if you insist on designing the thing
with the lowest common denominator as your basis, that's fine, but of
course the end result will be less than pretty unless designed *very*
carefully.

>Making software free, but only for folks with enough money
>to buy first class hardware is an interesting concept.

Except that Linux was designed more for the purposes of the designer than
anything else. If I were writing an OS, I'd design it to suit myself, too.
It's just that Linus was nice enough to share his code with the rest of us.

>Of course 5 years from now that will be different, but 5 years from now
>everyone will be running free GNU on their 200 MIPS, 64M SPARCstation-5.

Maybe. But by then, the 386/486 will probably be where the PC is now:
everyone will have one and they'll be dirt cheap. The timing will be
about right. In which case Linux will fit right in, wouldn't you say?

>>Re 2: your job is being a professor and researcher: That's one hell of a
>>good excuse for some of the brain-damages of minix. I can only hope (and
>>assume) that Amoeba doesn't suck like minix does.
>Amoeba was not designed to run on an 8088 with no hard disk.

Here's a question for you: as a general rule, when you go to design an
operating system, do you design it for specific capabilities and then run
it on whatever hardware will do the job, or do you design it with the
hardware as a target and fit the capabilities to the hardware? With respect
to Minix, it seems you did the latter, but I don't know whether or not you
did that with Amoeba.

>>If this was the only criterion for the "goodness" of a kernel, you'd be
>>right. What you don't mention is that minix doesn't do the micro-kernel
>>thing very well, and has problems with real multitasking (in the
>>kernel). If I had made an OS that had problems with a multithreading
>>filesystem, I wouldn't be so fast to condemn others: in fact, I'd do my
>>damndest to make others forget about the fiasco.
>A multithreaded file system is only a performance hack.

Bull. A multithreaded file system has a completely different design than
a single-threaded file system and has different design criteria than a
single-threaded file system.

>When there is only one job active, the normal case on a small PC, it buys
>you nothing and adds complexity to the code.

If there is only going to be one job active anyway then *why bother with
multitasking at all*????

If you're going to implement multitasking, then don't do a halfway job
of it. On the other hand, if you're going to assume that there will be
only one job active anyway, then don't bother with multitasking (after
all, it *does* complicate things :-).

>On machines fast enough to
>support multiple users, you probably have enough buffer cache to insure a
>hit cache hit rate, in which case multithreading also buys you nothing.

Maybe. Multiple users means multiple things being done simultaneously. I
wouldn't bet on the buffer cache buying you so much that multithreading
makes no difference. It's one thing if the users are doing something
simple, like editing a file. It's another thing if they're compiling,
reading news, or other things that touch lots of different files.

>It is only a win when there are multiple processes actually doing real disk
>I/O.

Which happens a *lot* when you're running multiple users. Or when you're
a machine hooked up to the net and handling news traffic.

>Whether it is worth making the system more complicated for this case is
>at least debatable.

Oh, come on. How tough is it to implement a multi-threaded file system?
All you need is a decent *buffered* (preferably infinitely so)
message-passing system and a way to save your current state when you send
out a request to the device driver(s) to perform some work (and obviously
some way to restore that state). Minix has the second via the setjmp()/
longjmp() mechanism, but lacks the former in a serious way.

>I still maintain the point that designing a monolithic kernel in 1991 is
>a fundamental error.

Not if you're trying to implement the system call semantics of Unix in a
reasonably simple and elegant way.

>Be thankful you are not my student. You would not
>get a high grade for such a design :-)

Why not? What's this big thing against monolithic kernels? There are
certain classes of problems for which a monolithic kernel is a more
appropriate design than a microkernel architecture. I think implementing
Unix semantics with a minimum of fuss is one such problem.

Unless you can suggest an elegant way to terminate a system call upon
receipt of a signal from within a microkernel OS?

>>The fact is that linux is more portable than minix. What? I hear you
>>say. It's true - but not in the sense that ast means: I made linux as
>>conformant to standards as I knew how (without having any POSIX standard
>>in front of me). Porting things to linux is generally /much/ easier
>>than porting them to minix.
>MINIX was designed before POSIX, and is now being (slowly) POSIXized as
>everyone who follows this newsgroup knows. Everyone agrees that user-level
>standards are a good idea. As an aside, I congratulate you for being able
>to write a POSIX-conformant system without having the POSIX standard in front
>of you. I find it difficult enough after studying the standard at great length.
>
>My point is that writing a new operating system that is closely tied to any
>particular piece of hardware, especially a weird one like the Intel line,
>is basically wrong.

Weird as the Intel line may be, it's *the* most popular line, by several
times. So it's not like it's *that* big a loss. And Intel hardware is
at least relatively cheap to come by, regardless of what your students
might tell you (why do you think they all own PCs?)...

>An OS itself should be easily portable to new hardware
>platforms.

As long as you don't sacrifice too much in the way of performance or
architectural elegance in order to gain this. Unfortunately, that's
*exactly* what happened with Minix: in attempting to implement it on
hardware of the lowest caliber, you ended up having to make design
decisions with respect to the architecture and implementation that have
made vintage Minix unusable as anything more than a personal toy operating
system. For example: why didn't you implement a system call server as
a layer between the file system and user programs? My guess: you didn't
have enough memory on the target machine to do it.

Put another way: you hit your original goal right on target, and are to
be applauded for that. But in doing so, you missed a lot of other
targets that wouldn't have been hard to hit as well, with some
consideration of them. I think. But I wasn't there when you were making
the decisions, so it's real hard for me to say for sure. I'm speaking
from hindsight, but you had the tough problem of figuring out what to do
without such benefit.

Now, *modified* Minix is usable. Add a bigger buffer cache. Modify it
so that it can take advantage of 386 protected mode. Fix the tty driver
so that it will give you multiple consoles. Fix the rs232 driver to deal
with DCD/DTR and do the right thing when carrier goes away. Fix the pipes
so that read and write requests don't fail just because they happen to be
bigger than the size of a physical pipe. Add shared text segments so you
maximize the use of your RAM. Fix the scheduler so that it deals with
character I/O bound processes in a reasonable way.

>When OS/360 was written in assembler for the IBM 360
>25 years ago, they probably could be excused. When MS-DOS was written
>specifically for the 8088 ten years ago, this was less than brilliant, as
>IBM and Microsoft now only too painfully realize.

Yeah, right. Just what hardware do you think they'd like to port DOS to,
anyway? I can't think of any. I don't think IBM or Microsoft are
regretting *that* particular aspect of DOS. Rather, they're probably
regretting the fact that it was written for the address space provided
by the 8088.

MS-DOS isn't less than brilliant because it was written for one machine
architecture. It's less than brilliant because it doesn't do anything
well, *regardless* of its portability or lack thereof.

>Writing a new OS only for the
>386 in 1991 gets you your second 'F' for this term. But if you do real well
>on the final exam, you can still pass the course.

He made his code freely redistributable. *You* didn't even do that. Just
for that move alone, he scores points in my book. Of course, the
distribution technology available to him is much better than what was
available when you did Minix, so it's hard to fault you for that...

But I must admit, Minix is still one hell of a bargain, and I would never
hesitate to recommend it to anyone who wants to learn something about Unix
and operating systems in general. As a working operating system (i.e.,
one intended for a multi-user environment), however, I'd hesitate to
recommend it, except that there really aren't any good alternatives
(except Linux, of course, at least tentatively. I can't say for sure,
since I haven't checked out Linux yet), since it doesn't have the performance
capabilities that a working operating system needs.

Kevin Brown

Linus Benedict Torvalds

unread,

Jan 31, 1992, 5:33:23 AM1/31/92

>The limitations of MINIX relate at least partly to my being a professor:
>An explicit design goal was to make it run on cheap hardware so students
>could afford it.

All right: a real technical point, and one that made some of my comments
inexcusable. But at the same time you shoot yourself in the foot a bit:
now you admit that some of the errors of minix were that it was too
portable: including machines that weren't really designed to run unix.
That assumption lead to the fact that minix now cannot easily be
extended to have things like paging, even for machines that would
support it. Yes, minix is portable, but you can rewrite that as
"doesn't use any features", and still be right.

>A multithreaded file system is only a performance hack.

Not true. It's a performance hack /on a microkernel/, but it's an
automatic feature when you write a monolithic kernel - one area where
microkernels don't work too well (as I pointed out in my personal mail
to ast). When writing a unix the "obsolete" way, you automatically get
a multithreaded kernel: every process does it's own job, and you don't
have to make ugly things like message queues to make it work
efficiently.

Besides, there are people who would consider "only a performance hack"
vital: unless you have a cray-3, I'd guess everybody gets tired of
waiting on the computer all the time. I know I did with minix (and yes,
I do with linux too, but it's /much/ better).

>I still maintain the point that designing a monolithic kernel in 1991 is
>a fundamental error. Be thankful you are not my student. You would not
>get a high grade for such a design :-)

Well, I probably won't get too good grades even without you: I had an
argument (completely unrelated - not even pertaining to OS's) with the
person here at the university that teaches OS design. I wonder when
I'll learn :)

>My point is that writing a new operating system that is closely tied to any
>particular piece of hardware, especially a weird one like the Intel line,
>is basically wrong.

But /my/ point is that the operating system /isn't/ tied to any
processor line: UNIX runs on most real processors in existence. Yes,
the /implementation/ is hardware-specific, but there's a HUGE
difference. You mention OS/360 and MS-DOG as examples of bad designs
as they were hardware-dependent, and I agree. But there's a big
difference between these and linux: linux API is portable (not due to my
clever design, but due to the fact that I decided to go for a fairly-
well-thought-out and tested OS: unix.)

If you write programs for linux today, you shouldn't have too many
surprises when you just recompile them for Hurd in the 21st century. As
has been noted (not only by me), the linux kernel is a miniscule part of
a complete system: Full sources for linux currently runs to about 200kB
compressed - full sources to a somewhat complete developement system is
at least 10MB compressed (and easily much, much more). And all of that
source is portable, except for this tiny kernel that you can (provably:
I did it) re-write totally from scratch in less than a year without
having /any/ prior knowledge.

In fact the /whole/ linux kernel is much smaller than the 386-dependent
things in mach: i386.tar.Z for the current version of mach is well over
800kB compressed (823391 bytes according to nic.funet.fi). Admittedly,
mach is "somewhat" bigger and has more features, but that should still
tell you something.

Linus

-Pete French.

unread,

Jan 31, 1992, 4:49:37 AM1/31/92

in article <1992Jan30....@epas.toronto.edu>, meg...@epas.utoronto.ca (David Megginson) says:
> Nntp-Posting-Host: epas.utoronto.ca

>
> In article 1992Jan30.1857...@netcom.COM feu...@netcom.COM (David Feustel) writes:
>>
>>That's ok. Einstein got lousy grades in math and physics.
>
> And Dan Quayle got low grades in political science. I think that there
> are more Dan Quayles than Einsteins out there... ;-)

What a horrible thought !

But on the points about microkernel v monolithic, isnt this partly an
artifact of the language being used ? MINIX may well be designed as a
microkernel system, but in the end you still end up with a large
monolithic chunk of binary data that gets loaded in as "the OS". Isnt it
written as separate programs simply because C does not support the idea
of multiple processes within a single piece of monolithic code. Is there
any real difference between a microkernel written as several pieces of C
and a monolithic kernel written in something like OCCAM ? I would have
thought that in this case the monolithic design would be a better one
than the micorkernel style since with the advantage of inbuilt
language concurrency the kernel could be made even more modular than the
MINIX one is.

Anyone for MINOX :-)

-bat.
--
-Pete French. (the -bat. ) /
Adaptive Systems Engineering /

Jyrki Kuoppala

unread,

Jan 31, 1992, 7:07:46 AM1/31/92

In article <12...@star.cs.vu.nl>, ast@cs (Andy Tanenbaum) writes:
>who want to turn MINIX in BSD UNIX off my back. But in all honesty, I would
>suggest that people who want a **MODERN** "free" OS look around for a
>microkernel-based, portable OS, like maybe GNU or something like that.

I hear bsd 4.4 might also become free and appear in the near future
for the 386, also someone's supposed to be working on bsd 4.4 on top
of the Mach microkernel, and then there's of course GNU. Currently of
course for many people Linux is the OS to use because it's here now,
is free and works.

>P.S. Just as a random aside, Amoeba has a UNIX emulator (running in user
>space), but it is far from complete. If there are any people who would
>like to work on that, please let me know. To run Amoeba you need a few 386s,
>one of which needs 16M, and all of which need the WD Ethernet card.

A note here, the sources I've seen seem to imply that Amoeba will not
be free as in you won't be able to use it, copy it, enhance it, share
it etc. without paying and/or asking permission from someone.

//Jyrki

Ari Lemmke

unread,

Jan 31, 1992, 6:38:50 PM1/31/92

: And reply I did, with complete abandon, and no thought for good taste
: and netiquette. Apologies to ast, and thanks to John Nall for a friendy
: "that's not how it's done"-letter. I over-reacted, and am now composing

I didn't and still don't see anything wrong to FOLLOUP, if
I'm getting *bashed* on the net. Linus' article was clear
and not against 'good taste' (what ever that is).

Linus doesn't have anything to apologise, not even on
comp.os.minix.

: Linus "my first, and hopefully last flamefest" Torvalds

arl // has nothing to do what I'm thinking about
// Minix or Linux.

Douglas Graham

unread,

Jan 31, 1992, 7:26:30 PM1/31/92

Can you recommend any (unbiased) literature that points out the strengths
and weaknesses of the two approaches? I'm sure that there is something
to be said for the microkernel approach, but I wonder how closely
Minix resembles the other systems that use it. Sure, Minix uses lots
of tasks and messages, but there must be more to a microkernel architecture
than that. I suspect that the Minix code is not split optimally into tasks.

> The only real argument for monolithic systems was performance, and there
> is now enough evidence showing that microkernel systems can be just as
> fast as monolithic systems (e.g., Rick Rashid has published papers comparing
> Mach 3.0 to monolithic systems) that it is now all over but the shoutin`.

My main complaint with Minix is not it's performance. It is that adding
features is a royal pain -- something that I presume a microkernel
architecure is supposed to alleviate.

> MINIX is a microkernel-based system.

Is there a consensus on this?

This is a fine assertion, but I've yet to see any rationale for it.
Linux is only about 12000 lines of code I think. I don't see how
splitting that into tasks and blasting messages around would improve it.

Well, there are no other choices that I'm aware of at the moment. But
when GNU OS comes out, I'll very likely jump ship again. I sense that
you *are* somewhat unhappy about Linux (and that surprises me somewhat).
I would guess that the reason so many people embraced it, is because it
offers more features. Your approach to people requesting features in
Minix, has generally been to tell them that they didn't really want that
feature anyway. I submit that the exodus in the direction of Linux
proves you wrong.

Disclaimer: I had nothing to do with Linux development. I just find
it an easier system to understand than Minix.
--
Doug Graham dgr...@bnr.ca My opinions are my own.

Charles Hedrick

unread,

Jan 31, 1992, 7:27:04 PM1/31/92

The history of software shows that availability wins out over
technical quality every time. That's Linux' major advantage. It's a
small 386-based system that's fairly compatible with generic Unix, and
is freely available. I dropped out of the Minix community a couple of
years ago when it became clear that (1) Minix was not going to take
advantage of anything beyond the 8086 anytime in the near future, and
(2) the licensing -- while amazingly friendly -- still made it hard
for people who were interested in producing a 386 version. Several
people apparently did nice work for the 386. But all they could
distribute were diffs. This made bringing up a 386 system a job that
isn't practical for a new user, and in fact I wasn't sure I wanted to
do it.

I apologize if things have changed in the last couple of years. If
it's now possible to get a 386 version in a form that's ready to run,
the community has developed a way to share Minix source, and bringing
up normal Unix programs has become easier in the interim, then I'm
willing to reconsider Minix. I do like its design.

It's possible that Linux will be overtaken by Gnu or a free BSD.
However, if the Gnu OS follows the example of all other Gnu software,
it will require a system with 128MB of memory and a 1GB disk to use.
There will still be room for a small system. My ideal OS would be 4.4
BSD. But 4.4's release date has a history of extreme slippage. With
most of their staff moving to BSDI, it's hard to believe that this
situation is going to be improved. For my own personal use, the BSDI
system will probably be great. But even their very attractive pricing
is likely to be too much for most of our students, and even though
users can get source from them, the fact that some of it is
proprietary will again mean that you can't just put altered code out
for public FTP. At any rate, Linux exists, and the rest of these
alternatives are vapor.

Theodore Y. Ts'o

unread,

Jan 31, 1992, 4:40:23 PM1/31/92

>From: a...@cs.vu.nl (Andy Tanenbaum)

>ftp.cs.vu.nl = 192.31.231.42 in dir minix/simulator.) I think it is a
>gross error to design an OS for any specific architecture, since that is
>not going to be around all that long.

It's not your fault for believing that Linux is tied to the 80386
architecture, since many Linux supporters (including Linus himself) have
made the this statement. However, the amount of 80386-specific code is
probably not much more than what is in a Minix implementation, and there
is certainly a lot less 80386 specific code in Linux than here is
Vax-specific code in BSD 4.3.

Granted, the port to other architectures hasn't been done yet. But if I
were going to bring up a Unix-like system on a new architecture, I'd
probably start with Linux rather than Minix, simply because I want to
have some control over what I can do with the resulting system when I'm
done with it. Yes, I'd have to rewrite large portions of the VM and
device driver layers --- but I'd have to do that with any other OS.
Maybe it would be a little bit harder than it would to port Minix to the
new architecture; but this would probably be only true for the first
architecture that we ported Linux to.

>While I could go into a long story here about the relative merits of the
>two designs, suffice it to say that among the people who actually design
>operating systems, the debate is essentially over. Microkernels have won.
>The only real argument for monolithic systems was performance, and there
>is now enough evidence showing that microkernel systems can be just as
>fast as monolithic systems (e.g., Rick Rashid has published papers comparing
>Mach 3.0 to monolithic systems) that it is now all over but the shoutin`.

This is not necessarily the case; I think you're painting a much more
black and white view of the universe than necessarily exists. I refer
you to such papers as Brent Welsh's (we...@parc.xerox.com) "The
Filsystem Belongs in the Kernel" paper, where in he argues that the
filesystem is a mature enough abstraction that it should live in the
kernel, not outside of it as it would in a strict microkernel design.

There also several people who have been concerned about the speed of
OSF/1 Mach when compared with monolithic systems; in particular, the
nubmer of context switches required to handle network traffic, and
networked filesystems in particular.

I am aware of the benefits of a micro kernel approach. However, the
fact remains that Linux is here, and GNU isn't --- and people have been
working on Hurd for a lot longer than Linus has been working on Linux.
Minix doesn't count because it's not free. :-)

I suspect that the balance of micro kernels versus monolithic kernels
depend on what you're doing. If you're interested in doing research, it
is obviously much easier to rip out and replace modules in a micro
kernel, and since only researchers write papers about operating systems,
ipso facto micro kernels must be the right approach. However, I do know
a lot of people who are not researchers, but who are rather practical
kernel programmers, who have a lot of concerns over the cost of copying
and the cost of context switches which are incurred in a micro kernel.

By the way, I don't buy your arguments that you don't need a
multi-threaded filesystem on a single user system. Once you bring up a
windowing system, and have a compile going in one window, a news reader
in another window, and UUCP/C News going in the background, you want
good filesystem performance, even on a single-user system. Maybe to a
theorist it's an unnecessary optimization and a (to use your words)
"performance hack", but I'm interested in a Real operating system ---
not a research toy.
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Theodore Ts'o bloom-beacon!mit-athena!tytso
308 High St., Medford, MA 02155 ty...@athena.mit.edu
Everybody's playing the game, but nobody's rules are the same!

j...@jshark.rn.com

unread,

Jan 31, 1992, 7:55:21 AM1/31/92

>In article <1992Jan29.1...@epas.toronto.edu> meg...@epas.utoronto.ca (David Megginson) writes:
>>
>>Why does the
>>Intel architecture _not_ allow drivers to be independent programs?
>
>The drivers have to read and write the device registers in I/O space, and
>this cannot be done in user mode on the 286 and 386. If it were possible
>to do I/O in a protected way in user space,

[[We must be talking about protected mode]] *THIS IS UNTRUE*

The Intel architecture supports independent tasks, each of which can be
given a "i/o privilege level". The convenient approach, used by iRMX(?), is
to "build" a load image ("root" device driver, kernel, MM and FS). Once
booted, these could be replaced by loadable tasks from disc (or network...)
and given a suitable privilege level.

The '386 additionally allows each task to have an "i/o permissions bitmap"
which specifies exactly which ports can be used.
(See "80386 Programmers Reference Manual", chapter 8)

> all the I/O tasks could have
>been user programs, like FS and MM.

Do you really mean "user programs" and not "separate tasks" ??

Separate tasks, possibly privileged, I'll agree with.

User level programs may be ok for teaching operating system principles, or on
toy computers :-) But a "production" system? Not on my machines!

>Andy Tanenbaum (a...@cs.vu.nl)

joe.
--
j...@jshark.rn.com
uunet!nstar!jshark!joe

j...@jshark.rn.com

unread,

Jan 31, 1992, 8:21:44 AM1/31/92

If you looked at the source instead of believing the author, you'd realise
this is not true!

He's replaced 'fubyte' by a routine which explicitly uses a segment register
- but that could be easily changed. Similarly, apart from a couple of places
which assume the '386 MMU, a couple of macros to hide the exact page sizes
etc would make porting trivial. Using '386 TSS's makes the code simpler,
but the VAX and WE32000 have similar structures.

As he's already admitted, a bit of planning would have the the system
neater, but merely putting '386 assembler around isn't a crime!

And with all due respect:
- the Book didn't make an issue of portability (apart from a few
"#ifdef M8088"s)
- by the time it was released, Minix had come to depend on several
8086 "features" that caused uproar from the 68000 users.

Will Rose

unread,

Feb 1, 1992, 7:16:12 AM2/1/92

I've used Minix quite a bit on a PC XT, from version 1.2 onwards, and
a couple of points seem worth making. Firstly that I ordered version
1.1 from Prentice Hall, and am devoutly thankful that they delayed my
order until 1.2 was available. The first version of something as
complicated as an OS is only for the dedicated, and that goes for Linux
too I should think.

Secondly Minix has evolved to a reliable OS on its original PC platform,
but is still getting there on eg. the Mac; these things do take time.

Thirdly even (standard) PC 1.5 Minix won't run a lot of current Unix
software. Partly this is a matter of the hardware being too limited,
and partly a matter of Minix being too limited in eg: the tty driver.
(And even this tty driver took a lot of sorting out in the early days).

Fourthly, I bought my XT four years ago - the motherboard was $110,
and memory (falling in price) was $7.00 per 256KB chip. Last autumn
I bought my wife an XT to replace her CP/M word-processor - the m/b
was 50,andmemorywas50, and memory was 50,andmemorywas1.50 a chip. This week I replaced a dead
286 board for a friend - the drop-in 16MHz 386SX was $140, and memory
was $40 for 9 x 1MB... If I actually wanted an OS to use today, I
think I'd go with Linux; but if I wanted to learn about OS's, I think
I'd use Minix. It looks as if they both do what they were designed
to do.

Will
c...@pnet01.cts.com

n.h.chandler

unread,

Feb 1, 1992, 8:38:51 PM2/1/92

I have been following the Minix/Linux discussion. How
can I get a copy of Linux?

Neville H. Chandler
cbnewsj!n...@att.com

Drew Eckhardt

unread,

Feb 2, 1992, 7:17:44 AM2/2/92

Every 386 TSS has an iopermission bitmap. If the CPL is of a lower priveledge
level than IOPL, the io permissions bitmap is consulted, allowing protection
on a port by port basis.

Allan Duncan

unread,

Feb 2, 1992, 5:06:26 PM2/2/92

From article <1992Jan30....@menudo.uh.edu>, by ke...@nuchat.sccsi.com (Kevin Brown):

> The *entire* system call interface in Minix needs to be rethought. As it
> stands right now, the file system is not just a file system, it's also a
> system-call server. That functionality needs to be separated out in order
> to facilitate a multiple file system architecture. Message passing is
> probably the right way to go about making the call and waiting for it, but
> the message should go to a system call server, not the file system itself.
>
> In order to handle all the special caveats of the Unix API, you end up writing
> a monolithic "kernel" even if you're using a microkernel base. You end up
> with something called a "server", and an example is the BSD server that runs
> under Mach.
>
> And, in any case, the message-passing in Minix needs to be completely redone.
> As it is, it's a kludge. I've been giving this some thought, but I haven't
> had time to do anything with what I've thought of so far. Suffice it to say
> that the proper way to do message-passing is probably with message ports
> (both public and private), with the various visible parts of the operating
> system having public message ports. Chances are, that ends up being the
> system call server only, though this will, of course, depend on the goals
> of the design.

It gets to sound more and more like Tripos and the Amiga :-)

Allan Duncan ACSnet adu...@trl.oz
(+613) 541 6708 Internet adu...@trl.oz.au
UUCP {uunet,hplabs,ukc}!munnari!aduncan
Telecom Research Labs, PO Box 249, Clayton, Victoria, 3168, Australia.

j...@jshark.rn.com

unread,

Feb 2, 1992, 6:59:12 PM2/2/92

>I was in the U.S. for a couple of weeks, so I haven't commented much on
>LINUX (not that I would have said much had I been around), but for what
>it is worth, I have a couple of comments now.

Maybe keepng quiet would have been best.

>1. MICROKERNEL VS MONOLITHIC SYSTEM
>

No, MS-DOS won. Sad, but there you are. 60 million: Next

It would be churlish to point out that MS-DOS has loadable device drivers
and that VMS is now (basically)a set of loadable service modules and drivers.

"Microkernel" was the buzz-word of last year, so Minix is a microkernel.
"Object-oriented" is this years, so Minix is object-oriented - right?

joe.
----
j...@jshark.rn.com
uunet!nstar!jshark!joe

I'm a mutated .sig virus, I got this from Henry Spencer's:
"As a user, I'll take speed over features anyday" - A Tanenbaum

Kevin Brown

unread,

Feb 3, 1992, 12:12:58 AM2/3/92

It has been brought to my attention that my last posting was exceedingly
harsh. Having reread it, I'm inclined to agree.

Dr. Tanenbaum claims that the microkernel architecture is the way to go.
He has a great deal more experience with operating systems than I have.
It's an understatement that it's likely that there's some substance to
his statement. :-)

Many of the things I said in my previous posting were more a result of my
philosophical viewpoint on operating systems and programming in general
than experience. And the particular viewpoint I hold that's relevent to
the discussion says that the method of implementation chosen depends on
the design goals, and that there is no "wrong" or "right" way to do things
that is independent of such goals. Thus, my statement that a monolithic
kernel follows from some design goals, e.g. ease of implementation of the
semantics of the Unix API. In particular, the ease of implementing things
like signal handling, premature system call termination, etc. At least,
that's the conclusion I come to when I think about the problem.

My experience with Minix says that there are a number of things that should
not go in a user process, things that are better left in the kernel. Things
like memory allocation (which requires global knowledge of the hardware,
something that a user process should, IMHO, not have) and signal handling
(which requires building stack frames).

чіSo from my point of view, the architecture of Minix is not ideal. While
it may win in that it's a "microkernel" architecture, the division of
functionality is not entirely to my liking. As is undoubtedly plainly
obvious by now. :-)

Despite that, Minix is quite usable in many ways as a personal operating
system, i.e. one where there is usually only one person logged into the
system. If I gave the impression that I thought it was unusable in general,
then I apologize for that.

However, as a *multiuser* operating system, ютi.e. an operating system designed
to efficiently meet the needs of multiple users simultaneously while also
performing batch operations, Minix is lacking, as far as I'm concerned.
The main reason, of course, is the single-threaded file system (hereafter,
STFS). Now, Dr. Tanenbaum may feel that a multi-threaded file system
(hereafter, MTFS) is merely a performance hack. Perhaps he's right.
Perhaps the architecture of a MTFS is sufficiently similar to that of a
STFS that his assessment is correct. My vision of a MTFS may differ
significantly from his, and this would explain why he and I seem to have
a difference of opinion on this matter. Regardless of whether or not a
MTFS is a "performance hack", for a *multiuser* operating system, I think
there are a lot of good arguments that say that a MTFS is a *necessary*
"performance hack". Provided, of course, that one does not have infinite
buffer cache resources. :-)

There are other things I feel Minix lacks as well. The ability to allocate
memory in the kernel is one (such an ability would allow any user process,
e.g. device drivers and the file system, to allocate memory dynamically.
This is useful for doing things like resizing the buffer cache on the fly,
etc. The ability to pass arbitrarily sized messages, optionally via shared
memory, is another (such an ability might be limited by constraints like
page size and such).

However much Minix may be lacking from my standpoint, it is nevertheless
a very useful and welcome enhancement to my system. In spite of the
impression that I may have given everyone in my last posting, there will
always be a soft spot in my heart for it, if only because it's the first
decent operating system I've had on my system that I've had source to.
I don't have to tell you people how incredibly useful it is to have source.
You already know.

It is very important to me to have source code to the things I run. It
bothers me a great deal to run things that I don't have source to. Even
the C compiler. And the less expensive the source is, the better. This
is why Dr. Tanenbaum's statements about Linux touched a raw nerve with me:
Linux comes with source *and* it's free. And it's available right now.

Someone, either here on this newsgroup or over on alt.os.linux, made a
very valid observation: the cost of a 16 MHz 386SX system is about $140
more than a comparably equipped (in terms of RAM size, display technology,
hard drive space, etc.) 8088 system. Minix is $169. In economic terms,
Linux wins if you have to buy Minix.

Where Minix wins (or is at least even :-) is when you can get it for free
via the educational distribution clause of the license agreement. However,
Minix will run even better on a 16 MHz 386SX than on an 8088. If I were
a student, I'd get the 386SX unless I simply didn't have a choice. Then
I'd get whichever operating system I could get for the least cost. If I
could get both for free, then I'd get both. :-)

Given the reasons Linus wrote Linux, I think it's hard for anyone to fault
him for writing it the way he did. And he was extremely nice in making
his code freely available to the rest of the world. It's not something he
had to do. In my book, that makes him almost beyond reproach.

Dr. Tanenbaum didn't make Minix free. His goals were different. Minix
is a teaching aid above all else (unless Dr. Tanenbaum has changed his
views about Minix :-). That means that he must be concerned with the
most efficient way to get Minix to the student population. At the time
Minix was released, Prentice-Hall was a good solution, and has been for
some time. However, I must wonder whether or not this is still the case.
Dr. Tanenbaum: do you still feel that free distribution of Minix via the
net is not the best way to distribute Minix?

Which wins? Minix or Linux? Depends on how you measure them...

Kevin Brown

peter da silva

unread,

Feb 3, 1992, 11:22:32 AM2/3/92

In article <1992Jan31....@ohm.york.ac.uk> pe...@ohm.york.ac.uk (-Pete French.) writes:
> But on the points about microkernel v monolithic, isnt this partly an
> artifact of the language being used ?

I doubt it.

[isn't MINIX]

> written as separate programs simply because C does not support the idea
> of multiple processes within a single piece of monolithic code.

C doesn't support formatted I/O either, but it can be implemented quite
effectively in C. So can concurrent processes. I've done it, in fact.
The resulting code is 90% portable (the 10% being the code that handles
the context switch).
--
-- Peter da Silva, Ferranti International Controls Corporation
-- Sugar Land, TX 77487-5012; +1 713 274 5180
-- "Have you hugged your wolf today?"

peter da silva

unread,

Feb 3, 1992, 11:37:24 AM2/3/92

Will you quit flaming each other?

I mean, linux is designed to provide a reasonably high performance environment
on a hardware platform crippled by years of backwards-compatible kludges. Minix
is designed as a teaching tool. Neither is that good at doing the other's job,
and why should they? The fact that Minix runs out of steam quickly (and it
does) isn't a problem in its chosen mileau. It's sure better than the TOY
operating system. The fact that Linux isn't transportable beyond the 386/AT
platform isn't a problem when there are millions of them out there (and quite
cheap: you can get a 386/SX for well under $1000).

A monolithic kernel is easy enough to build that it's worth doing it if it gets
a system out the door early. Think of it as a performance hack for programmer
time. The API is portable. You can replace the kernel with a microkernel
design (and MINIX isn't the be-all and end-all of microkernel designs either:
even for low end PCs... look at AmigaOS) without disturbing the applications.
That's the whole point of a portable API in the first place.

Microkernels are definitely a better design for many tasks. I takes more
work to make them efficient, so a simpler design that doesn't take advantage
of the microkernel in any real way is worth doing for pedagogical reasons.
Think of it as a performance hack for student time. The design is still good
and when you can get an API to the microkernel interface you can get VERY
impressive performance (thousands of context switches per second on an 8
MHz 68000).

peter da silva

unread,

Feb 3, 1992, 12:40:06 PM2/3/92

In article <1992Feb01.0...@bmerh2.bnr.ca> dgr...@bmers30.bnr.ca (Douglas Graham) writes:
> Minix resembles the other systems that use it. Sure, Minix uses lots
> of tasks and messages, but there must be more to a microkernel architecture
> than that. I suspect that the Minix code is not split optimally into tasks.

Definitely. Minix shows you how a microkernel works, but it sure doesn't show
you why you would use one.

A couple of years ago I brought this up with Andy, and his response indicated
that he was himself not convinced of the superiority of the microkernel design
at the time. He said (as near as I can recall... this is a paraphrase) that a
message passing design was inherently slower than a monolithic one... which was
news to me: I had (and still have) a PC that was MUCH more responsive than any
UNIX box I ever touched using a message-passing design.

> > MINIX is a microkernel-based system.

> Is there a consensus on this?

Yes, it's not a well-factored one, and there's no API to the microkernel
interface, but it's a microkernel design.

Richard Tobin

unread,

Feb 4, 1992, 9:46:49 AM2/4/92

>A multithreaded file system is only a performance hack. When there is only

>one job active, the normal case on a small PC, it buys you nothing

I find the single-threaded file system a serious pain when using
Minix. I often want to do something else while reading files from the
(excruciatingly slow) floppy disk. I rather like to play rogue while
waiting for large C or Lisp compilations. I look to look at files in
one editor buffer while compiling in another.

(The problem would be somewhat less if the file system stuck to
serving files and didn't interact with terminal i/o.)

Of course, in basic Minix with no virtual consoles and no chance of
running emacs, this isn't much of a problem. But to most people
that's a failure, not an advantage. It just isn't the case that on
single-user machines there's no use for more than one active process;
the idea only has any plausibility because so many people are used to
poor machines with poor operating systems.

As to portability, Minix only wins because of its limited ambitions.
If you wanted a full-featured Unix with paging, job-control, a window
system and so on, would it be quicker to start from basic Minix and
add the features, or to start from Linux and fix the 386-specific
bits? I don't think it's fair to criticise Linux when its aims are so
different from Minix's. If you want a system for pedagogical use,
Minix is the answer. But if what you want is an environment as much
like (say) a Sun as possible on your home computer, it has some
deficiencies.

-- Richard
--
Richard Tobin,
AI Applications Institute, R.T...@ed.ac.uk
Edinburgh University.

Ken Thompson

unread,

Feb 3, 1992, 6:07:54 PM2/3/92

viewpoint may be largely unrelated to its usefulness. Many if not
most of the software we use is probably obsolete according to the
latest design criteria. Most users could probably care less if the
internals of the operating system they use is obsolete. They are
rightly more interested in its performance and capabilities at the
user level.

I would generally agree that microkernels are probably the wave of
the future. However, it is in my opinion easier to implement a
monolithic kernel. It is also easier for it to turn into a mess in
a hurry as it is modified.

Regards,
Ken

--
Ken Thompson GTRI, Ga. Tech, Atlanta Ga. 30332 Internet:!k...@prism.gatech.edu
uucp:...!{allegra,amd,hplabs,ut-ngp}!gatech!prism!kt4
"Rowe's Rule: The odds are five to six that the light at the end of the
tunnel is the headlight of an oncoming train." -- Paul Dickson

Kevin Brown

unread,

Feb 4, 1992, 3:08:42 AM2/4/92

In article <47...@hydra.gatech.EDU> k...@prism.gatech.EDU (Ken Thompson) writes:
>viewpoint may be largely unrelated to its usefulness. Many if not
>most of the software we use is probably obsolete according to the
>latest design criteria. Most users could probably care less if the
>internals of the operating system they use is obsolete. They are
>rightly more interested in its performance and capabilities at the
>user level.
>
>I would generally agree that microkernels are probably the wave of
>the future. However, it is in my opinion easier to implement a
>monolithic kernel. It is also easier for it to turn into a mess in
>a hurry as it is modified.

How difficult is it to structure the source tree of a monolithic kernel
such that most modifications don't have a large negative impact on the
source? What sorts of pitfalls do you run into in this sort of endeavor,
and what suggestions do you have for dealing with them?

I guess what I'm asking is: how difficult is it to organize the source
such that most changes to the kernel remain localized in scope, even
though the kernel itself is monolithic?

I figure you've got years of experience with monolithic kernels :-),
so I'd think you'd have the best shot at answering questions like
these.

>Ken Thompson GTRI, Ga. Tech, Atlanta Ga. 30332 Internet:!k...@prism.gatech.edu
>uucp:...!{allegra,amd,hplabs,ut-ngp}!gatech!prism!kt4
>"Rowe's Rule: The odds are five to six that the light at the end of the
>tunnel is the headlight of an oncoming train." -- Paul Dickson

Kevin Brown

peter da silva

unread,

Feb 3, 1992, 12:32:54 PM2/3/92

In article <TYTSO.92J...@SOS.mit.edu> ty...@athena.mit.edu (Theodore Y. Ts'o) writes:
> This is not necessarily the case; I think you're painting a much more
> black and white view of the universe than necessarily exists. I refer
> you to such papers as Brent Welsh's (we...@parc.xerox.com) "The
> Filsystem Belongs in the Kernel" paper, where in he argues that the
> filesystem is a mature enough abstraction that it should live in the
> kernel, not outside of it as it would in a strict microkernel design.

What does "a mature enough abstraction" mean, here? Things don't move
into the kernel simply because they're now considered safe and stable
enough, but because they're too inefficient when they're outside it or
they lose functionality by being outside it, and there's no easy fix.

The Amiga operating system certainly benefits from having a file system
outside the kernel. There are dozens of file systems, many of them written
by hobbyists, available. Ideas like "assigned paths" can be played with
in the file system without breaking stuff. All these file systems have a
common interface and so look to the application as part of the operating
system, but just because something is on the other side of the API doesn't
mean it is, or belongs, in the kernel.

> There also several people who have been concerned about the speed of
> OSF/1 Mach when compared with monolithic systems; in particular, the
> nubmer of context switches required to handle network traffic, and
> networked filesystems in particular.

If this is because the networking was moved out of the kernel, I consider
it a price well worth paying. Having networking code in the kernel is the
source of many subtle bugs in networks. Just for something that bit us,
what happens if you need to get to the upper level driver before you can
acknowledge a packet, but the process that you need to run is hung up in
the tty driver waiting for a ^Q?

Something *I* would have expected to find in the kernel before now, yet
isn't, is windowing systems. With a microkernel (and the associated lower
*cost* of a context switch) you can get much of the advantages of a kernel
window system without paying the cost in complexity.

Kevin Brown

unread,

Feb 4, 1992, 3:28:08 AM2/4/92

In article <1992Feb2.2...@trl.oz.au> adu...@rhea.trl.OZ.AU (Allan Duncan) writes:
>From article <1992Jan30....@menudo.uh.edu>, by ke...@nuchat.sccsi.com (Kevin Brown):
>
>> The *entire* system call interface in Minix needs to be rethought. As it
>> stands right now, the file system is not just a file system, it's also a
>> system-call server. That functionality needs to be separated out in order
>> to facilitate a multiple file system architecture. Message passing is
>> probably the right way to go about making the call and waiting for it, but
>> the message should go to a system call server, not the file system itself.
>>
>> In order to handle all the special caveats of the Unix API, you end up writing
>> a monolithic "kernel" even if you're using a microkernel base. You end up
>> with something called a "server", and an example is the BSD server that runs
>> under Mach.
>>
>> And, in any case, the message-passing in Minix needs to be completely redone.
>> As it is, it's a kludge. I've been giving this some thought, but I haven't
>> had time to do anything with what I've thought of so far. Suffice it to say
>> that the proper way to do message-passing is probably with message ports
>> (both public and private), with the various visible parts of the operating
>> system having public message ports. Chances are, that ends up being the
>> system call server only, though this will, of course, depend on the goals
>> of the design.
>
>It gets to sound more and more like Tripos and the Amiga :-)

There's no question that many of my ideas spring from the architecture
of the Amiga's operating system. It's pretty impressive to see a
message-passing, multitasking operating system that operates as fast
as the Amiga's OS does on hardware that slow. They did a lot of things
right.

There are some ideas that, I think, are my own. Or, at least, that I've
developed independently. For example, if you have a message-passing
system that includes the option to transfer message memory ownership to the
target process, then it naturally follows that you can globally optimize the
use of your block cache by making your block cache global with respect
to *all* filesystems. The filesystem code requests blocks from the
block cache manager and tells the block cache manager what device driver
to call and what parameters to send it when flushing the block. The block
cache manager replies with a message that is the size of a block (or, if
you wish to allocate several at a time, several blocks). Since
ownership is transferred as a result of passing the message, the block
cache manager can allocate the memory itself, optionally flushing as
many blocks as it needs in order to free up enough to send to the caller.
The block cache manager is, of course, a user process. If the filesystem
code is written right, you can kill the block cache manager in order to
disable the block cache. The filesystem will simply do its thing
unbuffered. Makes for a slow system, but at least you can do it. You
can also change the behavior of the buffer cache by sending control
messages to the cache manager. Can you say "tunable parameters"? :-)

You could also accomplish this with some sort of shared memory, but this
would require semaphore control of the allocation list. You'd also have
to figure out a way to flush bits of the cache when needed (easy to do
if you're a monolithic kernel, but I'm referring to a microkernel) without
colliding with another process writing into the block. Semaphore control
of the individual blocks as well?

>Allan Duncan ACSnet adu...@trl.oz
>(+613) 541 6708 Internet adu...@trl.oz.au
> UUCP {uunet,hplabs,ukc}!munnari!aduncan
>Telecom Research Labs, PO Box 249, Clayton, Victoria, 3168, Australia.

Kevin Brown

Julien Maisonneuve

unread,

Feb 3, 1992, 12:10:14 PM2/3/92

I would like to second Kevin brown in most of his remarks.
I'll add a few user points :
- When ast states that FS multithreading is useless, it reminds me of the many
times I tried to let a job run in the background (like when reading an archive on
a floppy), it is just unusable, the & shell operator could even have been left
out.
- Most interesting utilities are not even compilable under Minix because of the
ATK compiler's incredible limits. Those were hardly understandable on a basic PC,
but become absurd on a 386. Every stupid DOS compiler has a large model (more
expensive, OK). I hate the 13 bit compress !
- The lack of Virtual Memory support prevents people studying this area to
experiment, and prevents users to use large programs. The strange design of the
MM also makes it hard to modify.

The problem is that even doing exploratory work under minix is painful.
If you want to get any work done (or even fun), even DOS is becoming a better
alternative (with things like DJ GPP).
In its basic form, it is really no more than OS course example, a good
toy, but a toy. Obtaining and applying patches is a pain, and precludes further
upgrades.

Too bad when not so much is missing to make it really good.
Thanks for the work andy, but Linux didn't deserve your answer.
For the common people, it does many things better than Minix.

Julien Maisonneuve.

This is not a flame, just my experience.

Michael L. Kaufman

unread,

Feb 3, 1992, 5:27:48 PM2/3/92

I tried to send these two posts from work, but I think they got eaten. If you
have seen them already, sorry.

-------------------------------------------------------------------------------

Andy Tanenbaum writes an interesting article (also interesting was finding out
that he actually reads this group) but I think he is missing an important
point.

He Wrote:
>As most of you know, for me MINIX is a hobby, ...

Which is also probably true of most, if not all, of the people who are involved
in Linux. We are not developing a system to take over the OS market, we are
just having a good time.

> What is going to happen
> is that they will gradually take over from the 80x86 line. They will
> run old MS-DOS programs by interpreting the 80386 in software.

Well when this happens, if I still want to play with Linux, I can just run it
on my 386 simulator.

That's fine for the people who have those machines, but it wasn't a free
lunch. That portibility was gained at the cost of some performance and some
features on the 386. Before you decide that LINUX is not the way to go, you
should think about what it is going to be used for. I am going to use it for
running memory and computation intensive graphics programs on my 486. For me,
speed and memory were more important then future state-of-the-artness and
portability.

>But in all honesty, I would
>suggest that people who want a **MODERN** "free" OS look around for a
>microkernel-based, portable OS, like maybe GNU or something like that.

I don't know of any free microkernel-based, portable OSes. GNU is still
vaporware, and likely to remain that way for the forseeable future. Do
you actually have one to recomend, or are you just toying with me? ;-)

------------------------------------------------------------------------------

In article <12...@star.cs.vu.nl> a...@cs.vu.nl (Andy Tanenbaum) writes:
>My point is that writing a new operating system that is closely tied to any
>particular piece of hardware, especially a weird one like the Intel line,

>is basically wrong. An OS itself should be easily portable to new hardware
>platforms.

I think I see where I disagree with you now. You are looking at OS design
as an end in itself. Minix is good because it is portable/Micro-Kernal/etc.
Linux is not good because it is monolithic/tightly tied to Intel/etc. That
is not a strange attitude for someone in the acedemic world, but it is not
something you should expect to be universally shared. Linux is not being written
as a teaching tool, or as an abstract exercise. It is being written to allow
people to run GNU-type software _today_. The fact that it may not be in use
in five years is less important then the fact that today (well, by April
probably) I can run all sorts of software on it that I want to run. You keep
saying that Minix is better, but if it will not run the software that I want
to run, it really isn't that good (for me) at all.

> When OS/360 was written in assembler for the IBM 360
>25 years ago, they probably could be excused. When MS-DOS was written
>specifically for the 8088 ten years ago, this was less than brilliant, as
>IBM and Microsoft now only too painfully realize.

Same point. MSoft did not come out with Dos to "explore the frontiers of os
research". They did it to make a buck. And considering the fact that MS-DOS
probably still outsells everyone else put together, I don't think that you
say that they have failed _in their goals_. Not that MS-Dos is the best OS
in terms of anything else, only that it has served their needs.

Michael

--
Michael Kaufman | I've seen things you people wouldn't believe. Attack ships on
kaufman | fire off the shoulder of Orion. I watched C-beams glitter in
@eecs.nwu.edu | the dark near the Tannhauser gate. All those moments will be
| lost in time - like tears in rain. Time to die. Roy Batty

Jonathan Allen

unread,

Feb 3, 1992, 2:43:00 AM2/3/92

In article <12...@star.cs.vu.nl>, a...@cs.vu.nl (Andy Tanenbaum) wrote:
> In article <1992Jan29.1...@epas.toronto.edu> meg...@epas.utoronto.ca (David Megginson) writes:
>>
>>Why does the
>>Intel architecture _not_ allow drivers to be independent programs?
>
> The drivers have to read and write the device registers in I/O space, and
> this cannot be done in user mode on the 286 and 386. If it were possible
> to do I/O in a protected way in user space, all the I/O tasks could have
> been user programs, like FS and MM.

Surely this could have been done by a minute task just to read/write a
given port address in one message ? The security could have been checked
like everything else using the process table ...

Sure it would not have been at all efficient, but would have given
the independance at a price.

Jonathan

Andy Tanenbaum

unread,

Feb 5, 1992, 9:48:48 AM2/5/92

In article <61...@skye.ed.ac.uk> ric...@aiai.UUCP (Richard Tobin) writes:
>If you wanted a full-featured Unix with paging, job-control, a window
>system and so on, would it be quicker to start from basic Minix and
>add the features, or to start from Linux and fix the 386-specific
>bits?

Another option that seems to be totally forgotten here is buy UNIX or a
clone. If you just want to USE the system, instead of hacking on its
internals, you don't need source code. Coherent is only $99, and there
are various true UNIX systems with more features for more money. For the
true hacker, not having source code is fatal, but for people who just
want a UNIX system, there are many alternatives (albeit not free).

Andy Tanenbaum (a...@cs.vul.nl)

Richard Tobin

unread,

Feb 5, 1992, 11🔞16 AM2/5/92

In article <1992Feb2.1...@colorado.edu> dr...@anchor.cs.colorado.edu (Drew Eckhardt) writes:
>Every 386 TSS has an iopermission bitmap. If the CPL is of a lower priveledge
>level than IOPL, the io permissions bitmap is consulted, allowing protection
>on a port by port basis.

I was looking into using this recently under Minix 386, and to check I
was doing the right thing, I wrote a user program to access the video
registers. The idea was to have it fail, and then change the kernel
to make it work. To my surprise, it worked anyway...

I'll take a closer look sometime, but does anyone (Bruce?) happen to
already know the explanation?

John W. Linville

unread,

Feb 5, 1992, 12:56:35 PM2/5/92

Coherent is limited by a compiler that only supports the small memory model,
making it just as difficult (perhaps more in some instances) to port 'standard'
Unix programs to Coherent as it can be under Minix. Also, Coherent is not
portable (or at least, to the best of my knowledge, has not been ported), so
this advocacy contradicts one of your arguments against Linux.

Since a true Unix system often costs as much as the machine it runs on (even
more since many Unix providers un-bundle networking and development packages),
buying a true Unix system is more than beyond the budget of many people.

John W. Linville

Lawrence C. Foard

unread,

Feb 5, 1992, 9:56:30 AM2/5/92

>Don`t get me wrong, I am not unhappy with LINUX. It will get all the people

>who want to turn MINIX in BSD UNIX off my back. But in all honesty, I would

>suggest that people who want a **MODERN** "free" OS look around for a
>microkernel-based, portable OS, like maybe GNU or something like that.

I believe you have some valid points, although I am not sure that a
microkernel is necessarily better. It might make more sense to allow some
combination of the two. As part of the IPC code I'm writting for Linux I am
going to include code that will allow device drivers and file systems to run
as user processes. These will be significantly slower though, and I believe it
would be a mistake to move everything outside the kernel (TCP/IP will be
internal).

Actually my main problem with OS theorists is that they have never tested
there ideas! None of these ideas (with a partial exception for MACH) has ever
seen the light of day. 32 bit home computers have been available for almost a
decade and Linus was the first person to ever write a working OS for them
that can be used without paying AT&T $100,000. A piece of software in hand is
worth ten pieces of vaporware, OS theorists are quick to jump all over an OS
but they are unwilling to ever provide an alternative.

The general consensus that Micro kernels is the way to go means nothing when
a real application has never even run on one.

The release of Linux is allowing me to try some ideas I've been wanting to
experment with for years, but I have never had the opportunity to work with
source code for a functioning OS.
--
Disclaimer: Opinions are based on logic rather than biblical "fact". ------
Hackers do it for fun. | First they came for the drug users, I said \ /
"Profesionals" do it for money. | nothing, then they came for hackers, \ /
Managers have others do it for them. | I said nothing... STOP W.O.D. \/

David Megginson

unread,

Feb 5, 1992, 3:50:55 PM2/5/92

>Another option that seems to be totally forgotten here is buy UNIX or a
>clone. If you just want to USE the system, instead of hacking on its
>internals, you don't need source code. Coherent is only $99, and there
>are various true UNIX systems with more features for more money. For the
>true hacker, not having source code is fatal, but for people who just
>want a UNIX system, there are many alternatives (albeit not free).

What Unix's _are_ available for a simple, M68000-based ST, with _or_
without source? These are the only options I know of:

OS 9.
The Beckmeyer MT C-Shell.
MiNT.
Minix.

I have used all of these except for OS 9, and Minix is clearly the
closest thing to Unix that I can run (though it is easier to port BSD
programs to MiNT using the MiNT gcc library). I could shell out CAN
$3000 for a TT, but then I may as well buy a 386 box anyway. Besides,
I _like_ having the source. The extra advantage of Minix is that the
user base is a lot wider than the ST market, so I can get decent
system enhancements from Amiga, Mac, Sparc, XT, AT, '386 and '486
users as well as from fellow ST owners.

David

Andy Tanenbaum

unread,

Feb 5, 1992, 6:33:23 PM2/5/92

I'm mortally insulted. I AM NOT A THEORIST. Ask anybody who was at our
department meeting yesterday (in joke).

Actually, these ideas have been very well tested in practice. OSF is betting
its whole business on a microkernel (Mach 3.0). USL is betting its business
on another one (Chorus). Both of these run lots of software, and both have
been extensively compared to monolithic systems. Amoeba has been fully
implemented and tested for a number of applications. QNX is a microkernel
based system, and someone just told me the installed base is 200,000 systems.
Microkernels are not a pipe dream. They represent proven technology.

The Mach guys wrote a paper called "UNIX as an application program."
It was by Golub et al., in the Summer 1990 USENIX conference. The Chorus
people also have a technical report on microkernel performance, and I
coauthored another paper on the subject, which I mentioned yesterday
(Dec. 1991 Computing Systems). Check them out.

Andy Tanenbaum (a...@cs.vu.nl)

Lawrence C. Foard

unread,

Feb 6, 1992, 4:22:40 AM2/6/92

In article <1992Feb3.0...@menudo.uh.edu> ke...@taronga.taronga.com (Kevin Brown) writes:
>Dr. Tanenbaum claims that the microkernel architecture is the way to go.
>He has a great deal more experience with operating systems than I have.
>It's an understatement that it's likely that there's some substance to
>his statement. :-)

I tend to prefer seeing for myself rather than accepting "expert" opinion.
Microkernels are nice asthetically, but there are times when practical issues
must also be considered :)

>w3So from my point of view, the architecture of Minix is not ideal. While

>it may win in that it's a "microkernel" architecture, the division of
>functionality is not entirely to my liking. As is undoubtedly plainly
>obvious by now. :-)

I've been told by people who have used both that Linux is significantly
faster. There are certainly several factors involved (certainly using 32 bits
helps alot), but the multithreading also makes for much lower overhead.

>However, as a *multiuser* operating system, ~ri.e. an operating system designed

>to efficiently meet the needs of multiple users simultaneously while also
>performing batch operations, Minix is lacking, as far as I'm concerned.
>The main reason, of course, is the single-threaded file system (hereafter,
>STFS). Now, Dr. Tanenbaum may feel that a multi-threaded file system
>(hereafter, MTFS) is merely a performance hack.

I think this is a very valid problem. There are two ways a single threaded FS
could work and both have substantial problems. If the FS blocks while waiting
for I/O it would be completely unusable for "real" work. Imagine several users
accessing a database, if the FS blocks for I/O they will have to wait
eventhough the data they are looking for is already in the cache. If it is
designed to be non blocking then it is even more complicated than a
multithreaded FS and will have more overhead. I hope it is atleast the second

>However much Minix may be lacking from my standpoint, it is nevertheless
>a very useful and welcome enhancement to my system. In spite of the
>impression that I may have given everyone in my last posting, there will
>always be a soft spot in my heart for it, if only because it's the first
>decent operating system I've had on my system that I've had source to.
>I don't have to tell you people how incredibly useful it is to have source.
>You already know.

I will agree here, Minix is infinitly better than Messy-Loss :)

>Given the reasons Linus wrote Linux, I think it's hard for anyone to fault
>him for writing it the way he did. And he was extremely nice in making
>his code freely available to the rest of the world. It's not something he
>had to do. In my book, that makes him almost beyond reproach.

I think more effort has been put into making practical use of Linux possible.
An educational OS is nice, but there is a world outside of colleges that
is suffering from the lack of cheap and useful OS's, I've been stuck doing
most consulting work in Messy Loss because customers don't want to fork out
$1000 for UNIX.

>Dr. Tanenbaum didn't make Minix free. His goals were different. Minix
>is a teaching aid above all else (unless Dr. Tanenbaum has changed his
>views about Minix :-). That means that he must be concerned with the
>most efficient way to get Minix to the student population. At the time
>Minix was released, Prentice-Hall was a good solution, and has been for
>some time. However, I must wonder whether or not this is still the case.
>Dr. Tanenbaum: do you still feel that free distribution of Minix via the
>net is not the best way to distribute Minix?

I would guess that Prentice-Hall would have some objections :)

--
Disclaimer: Opinions are based on logic rather than biblical "fact". ------

This is your friendly | First they came for the drug users, I said \ /
neighborhood signature virus | nothing, then they came for hackers, \ /
please add me to your signature! | I said nothing... STOP W.O.D. \/

Timothy Murphy

unread,

Feb 6, 1992, 6:14:59 AM2/6/92

>32 bit home computers have been available for almost a
>decade and Linus was the first person to ever write a working OS for them
>that can be used without paying AT&T $100,000. A piece of software in hand is
>worth ten pieces of vaporware, OS theorists are quick to jump all over an OS
>but they are unwilling to ever provide an alternative.

Surely Bruce Evans' 386-Minix preceded Linux?

(Diffs for PC-Minix -> 386-Minix
available from archive...@plains.nodak.edu
in the directory Minix/oz)

--
Timothy Murphy
e-mail: t...@maths.tcd.ie
tel: +353-1-2842366
s-mail: School of Mathematics, Trinity College, Dublin 2, Ireland

Tony Travis

unread,

Feb 5, 1992, 9:17:13 PM2/5/92

a...@cs.vu.nl (Andy Tanenbaum) writes:
> Another option that seems to be totally forgotten here is buy UNIX or a
> clone. If you just want to USE the system, instead of hacking on its
> internals, you don't need source code. Coherent is only $99, and there
> are various true UNIX systems with more features for more money. For the
> true hacker, not having source code is fatal, but for people who just
> want a UNIX system, there are many alternatives (albeit not free).

Andy, I have followed the development of Minix since the first messages
were posted to this group and I am now running 1.5.10 with Bruce
Evans's patches for the 386.

I 'just' want a Unix on my PC and I am not interested in hacking on its
internals, but I *do* want the source code!

An important principle underlying the success and popularity of Unix is
the philosophy of building on the work of others.

This philosophy relies upon the availability of the source code in
order that it can be examined, modified and re-used in new software.

Many years ago, I was in the happy position of being an AT&T Seventh
Edition Unix source licencee but, even then, I saw your decision to
make the source of Minix available as liberation from the shackles of
AT&T copyright!!

I think you may sometimes forget that your 'hobby' has had a profound
effect on the availability of 'personal' Unix (ie. affordable Unix) and
that the 8086 PC I ran Minix 1.2 on actually cost me considerably more
than my present 386/SX clone.

Clearly, Minix _cannot_ be all things to all men, but I see the
progress to 386 versions in much the same way that I see 68000 or other
linear address space architectures: it is a good thing for people like
me who use Minix and feel constrained by the segmented architecture of
the PC version for applications.

NOTHING you can say would convince me that I should use Coherent ...

Tony

--
-------------------------------------------------------------------------------
Dr. A.J.Travis a...@uk.ac.sari.rri | Rowett Research Institute,
| Greenburn Road, Bucksburn, Aberdeen,
| AB2 9SB. UK. tel 0224-712751

Jerry Shekhel

unread,

Feb 6, 1992, 4:28:05 PM2/6/92

linv...@garfield.catt.ncsu.edu (John W. Linville) writes:
>
>Since a true Unix system often costs as much as the machine it runs on (even
>more since many Unix providers un-bundle networking and development packages),
>buying a true Unix system is more than beyond the budget of many people.
>

For those who may be interested, MST sells System V Release 4.0.3 for the
386/486 for 399includingdevelopmentsystem,399 including development system, 399includingdevelopmentsystem,499 if you need networking.
X11R5 binaries may be obtained via FTP (networking is not required for X11R5).
I have just such a setup, and it works great. MST's version of UNIX doesn't
have too much in the way of bug fixes relative to the AT&T code, but the
only thing I've really had problems with was a couple of bugs in csh. Now
that I have tcsh working (built without so much as a warning!) I'll never go
back :-)

Micro Station Technology
1140 Kentwood Avenue
Cupertine, CA 95014
Tel: 408-253-3898
Fax: 408-253-7853

I am not affiliated with MST except as a customer.

peter da silva

unread,

Feb 6, 1992, 11:02:47 AM2/6/92

> QNX is a microkernel
> based system, and someone just told me the installed base is 200,000 systems.

Oh yes, while I'm on the subject... there are over 3 million Amigas out there,
which means that there are more of them than any UNIX vendor has shipped, and
probably more than all UNIX systems combined.

peter da silva

unread,

Feb 6, 1992, 11:00:22 AM2/6/92

I beg to differ... there are many microkernel operating systems out there
for everything from an 8088 (QNX) up to large research systems.

> None of these ideas (with a partial exception for MACH) has ever
> seen the light of day. 32 bit home computers have been available for almost a
> decade and Linus was the first person to ever write a working OS for them
> that can be used without paying AT&T $100,000.

I must have been imagining AmigaOS, then. I've been using a figment of my
imagination for the past 6 years.

AmigaOS is a microkernel message-passing design, with better response time
and performance than any other readily available PC operating system: including
MINIX, OS/2, Windows, MacOS, Linux, UNIX, and *certainly* MS-DOS.

The microkernel design has proven invaluable. Things like new file systems
that are normally available only from the vendor are hobbyist products on
the Amiga. Device drivers are simply shared libraries and tasks with specific
entry points and message ports. So are file systems, the window system, and
so on. It's a WONDERFUL design, and validates everything that people have
been saying about microkernels. Yes, it takes more work to get them off the
ground than a coroutine based macrokernel like UNIX, but the versatility
pays you back many times over.

I really wish Andy would do a new MINIX based on what has been learned since
the first release. The factoring of responsibilities in MINIX is fairly poor,
but the basic concept is good.

> The general consensus that Micro kernels is the way to go means nothing when
> a real application has never even run on one.

I'm dreaming again. I sure throught Deluxe Paint, Sculpt 3d, Photon Paint,
Manx C, Manx SDB, Perfect Sound, Videoscape 3d, and the other programs I
bought for my Amiga were "real". I'll have to send the damn things back now,
I guess.

The availability of Linux is great. I'm delighted it exists. I'm sure that
the macrokernel design is one reason it has been implemented so fast, and this
is a valid reason to use macrokernels. BUT... this doesn't mean that
microkernels are inherently slow, or simply research toys.

Tim W Smith

unread,

Feb 6, 1992, 8:30:59 PM2/6/92

Andy Tanenbaum (a...@cs.vu.nl) writes:
> The drivers have to read and write the device registers in I/O space, and
> this cannot be done in user mode on the 286 and 386. If it were possible
> to do I/O in a protected way in user space, all the I/O tasks could have
> been user programs, like FS and MM.

On the 386, you could run the drivers in V86 mode, which sort of counts
as user mode and allows access to I/O registers if the kernel sets things
up to allow this.

Tim Smith

Tim W Smith

unread,

Feb 6, 1992, 9:09:23 PM2/6/92

> Actually my main problem with OS theorists is that they have never tested
> there ideas! None of these ideas (with a partial exception for MACH) has ever
> seen the light of day. 32 bit home computers have been available for almost a
> decade and Linus was the first person to ever write a working OS for them
> that can be used without paying AT&T $100,000. A piece of software in hand is

How about Netware 386 from Novell? It seems to work.

Tim Smith

Richard Tobin

unread,

Feb 7, 1992, 9:58:22 AM2/7/92

>If you just want to USE the system, instead of hacking on its
>internals, you don't need source code.

Unfortunately hacking on the internals is just what many of us want
the system for... You'll be rid of most of us when BSD-detox or GNU
comes out, which should happen in the next few months (yeah, right).

bert thompson

unread,

Feb 8, 1992, 2:43:54 AM2/8/92

feu...@netcom.COM (David Feustel) writes:

>That's ok. Einstein got lousy grades in math and physics.

no he didn't.

bert.

Rogier Wolff

unread,

Feb 8, 1992, 4:13:39 AM2/8/92

a...@cs.vu.nl (Andy Tanenbaum) writes:

>In article <1992Feb5....@wpi.WPI.EDU> ent...@wintermute.WPI.EDU (Lawrence C. Foard) writes:
>>Actually my main problem with OS theorists is that they have never tested
>>there ideas!
>I'm mortally insulted. I AM NOT A THEORIST. Ask anybody who was at our
>department meeting yesterday (in joke).

>Actually, these ideas have been very well tested in practice.

The problem is that to really do an unbiased test you would need two
*identical* teams, and ask them to make two OS's, for the same
destination machine, one using a microkernel architecture, and the other
using the monolithic approach. This is in practice not feasable and the
publications on the subject can only shout: "look: I've got a good
performance using a microkernel", "we've got very good performance using
a monolithic aproach" or "it only took us X months to implement this OS"

If people did benchmark their OS's they wrote the OS for one architecture,
and adapted it to test the other. This adaptation will naturally degrade
performance, and show that the designers were right in the first place.

Anyway, anybody have an opinion about the fact that code for printf
is included three times in the Minix OS when it runs (once in the
kernel, MM and FS)

Roger

--
If the opposite of "pro" is "con", what is the opposite of "progress"?
(stolen from kadokev@iitvax ==? tech...@iitmax.iit.edu)
EMail: wo...@duteca.et.tudelft.nl ** Tel +31-15-783644 or +31-15-142371

David Megginson

unread,

Feb 8, 1992, 10:04:31 AM2/8/92

>Anyway, anybody have an opinion about the fact that code for printf
>is included three times in the Minix OS when it runs (once in the
>kernel, MM and FS)

Back in the yore days, this might have been a problem. I remember when
every program, even wordprocessors, had to be written in assembler to
squeeze them down to the smallest size possible for a 64K system. One
of the reasons WordPerfect is such a mess today is that it was written
in assembler instead of C.

Now, even the small systems which Minix runs have at least 640K, so a
few wasted bytes are not so much of a problem.

Why not write Linux in 80386 assembler? It would be smaller and even
faster. And don't forget to code inline as much as possible, to avoid
the crippling overhead of function calls. And leave out comments,
because they waste disk space.

Kevin Brown

unread,

Feb 9, 1992, 4:02:24 AM2/9/92

In article <1992Feb6.0...@wpi.WPI.EDU> ent...@wintermute.WPI.EDU (Lawrence C. Foard) writes:
>In article <1992Feb3.0...@menudo.uh.edu> ke...@taronga.taronga.com (Kevin Brown) writes:
>>Dr. Tanenbaum claims that the microkernel architecture is the way to go.
>>He has a great deal more experience with operating systems than I have.
>>It's an understatement that it's likely that there's some substance to
>>his statement. :-)
>
>I tend to prefer seeing for myself rather than accepting "expert" opinion.
>Microkernels are nice asthetically, but there are times when practical issues
>must also be considered :)

I agree. This is why I qualified my statement the way I did. :-)

Having seen both monolithic and microkernel architectures running, though,
I tend to agree that microkernels are generally the way to go, all other
things being equal.

But as you say, all things are not always equal. That's when it becomes
a judgement call. Which is better? Depends on what you're trying to do.

>I've been told by people who have used both that Linux is significantly
>faster. There are certainly several factors involved (certainly using 32 bits
>helps alot), but the multithreading also makes for much lower overhead.

Yup. I think that if Minix were arranged so that it had message queueing
and a true multithreaded filesystem, it might be comparable to a monolithic
kernel in terms of speed. It's hard for me to say, though. I haven't
played around much with multithreaded filesystems, so I don't know how
hard it is to make them work efficiently. I'd think, though, that it would
depend enormously on how efficient your device drivers were, and how much
data copying you'd have to do (ideally, you'd pass references to the data
buffers around and do your actual data transfers directly to the user's
buffer).

>>However, as a *multiuser* operating system, i.e. an operating system designed

>>to efficiently meet the needs of multiple users simultaneously while also
>>performing batch operations, Minix is lacking, as far as I'm concerned.
>>The main reason, of course, is the single-threaded file system (hereafter,
>>STFS). Now, Dr. Tanenbaum may feel that a multi-threaded file system
>>(hereafter, MTFS) is merely a performance hack.
>
>I think this is a very valid problem. There are two ways a single threaded FS
>could work and both have substantial problems. If the FS blocks while waiting
>for I/O it would be completely unusable for "real" work. Imagine several users
>accessing a database, if the FS blocks for I/O they will have to wait
>eventhough the data they are looking for is already in the cache. If it is
>designed to be non blocking then it is even more complicated than a
>multithreaded FS and will have more overhead. I hope it is atleast the second

I haven't gone deeply into the source code of the Minix file system, but
the impression I get from my perusing of it is that it blocks on disk I/O
but not on terminal I/O, the idea being that disk I/O requests will almost
always be satisfied relatively soon after they are made, whereas terminal
I/O requests can take an indefinite amount of time to satisfy.

But it seems to me that if you're going to implement the mechanism to handle
I/O where the file system doesn't block waiting for it, why not use that
mechanism universally???

>>However much Minix may be lacking from my standpoint, it is nevertheless
>>a very useful and welcome enhancement to my system. In spite of the
>>impression that I may have given everyone in my last posting, there will
>>always be a soft spot in my heart for it, if only because it's the first
>>decent operating system I've had on my system that I've had source to.
>>I don't have to tell you people how incredibly useful it is to have source.
>>You already know.
>
>I will agree here, Minix is infinitly better than Messy-Loss :)

Which is why I try to avoid using MS-DOS whenever possible. I'll bet a
lot of us Minixers do the same. :-)

>>Given the reasons Linus wrote Linux, I think it's hard for anyone to fault
>>him for writing it the way he did. And he was extremely nice in making
>>his code freely available to the rest of the world. It's not something he
>>had to do. In my book, that makes him almost beyond reproach.
>
>I think more effort has been put into making practical use of Linux possible.
>An educational OS is nice, but there is a world outside of colleges that
>is suffering from the lack of cheap and useful OS's, I've been stuck doing
>most consulting work in Messy Loss because customers don't want to fork out
>$1000 for UNIX.

Even students can make good use of something like Linux. I have 8 megabytes
of RAM on my machine, and 410 meg of harddrive space. Yet I can barely
run SBProlog on my system, even though my system is considerably more macho
than most. If I had demand paging on my system, this wouldn't be a problem,
but the only patches I have for demand paging seem not to work very well.
Once Linux becomes more stable (and gets support for Seagate ST-02 SCSI),
I'll snag the sources and check it out. Since I already own Minix, I can
legally transport *everything* over to it, and since both share the same
filesystem layout, I can do the transporting with a minimum of hassle.

>>Dr. Tanenbaum didn't make Minix free. His goals were different. Minix
>>is a teaching aid above all else (unless Dr. Tanenbaum has changed his
>>views about Minix :-). That means that he must be concerned with the
>>most efficient way to get Minix to the student population. At the time
>>Minix was released, Prentice-Hall was a good solution, and has been for
>>some time. However, I must wonder whether or not this is still the case.
>>Dr. Tanenbaum: do you still feel that free distribution of Minix via the
>>net is not the best way to distribute Minix?
>
>I would guess that Prentice-Hall would have some objections :)

No doubt. :-(

--
Kevin Brown Disclaimer: huh?
ke...@taronga.com ke...@nuchat.sccsi.com

peter da silva

unread,

Feb 9, 1992, 10:10:00 PM2/9/92

> The problem is that to really do an unbiased test you would need two

> *identical* teams, and ask them to make two OS's [...]

No, you don't. I don't think there's any question that a macrokernel is
very easy to get decent performance out of. Where the microkernel design
has a major advantage is in flexibility. Adding stuff to a macrokernel
is fairly complex and quickly becomes pretty gross. Look at BSD or System V
for examples. Adding stuff to a well designed microkernel is VERY easy.

Sometimes you don't want to compare oranges and oranges. Sometimes you want
to compare concentrated orange juice with fresh-squeezed. Fresh-squeezed
takes longer, but it's worth it.

Plus, with a microkernel you can get much better context switching between
microtasks than macro processes. So you can do stuff in separate processes
that would be out of the question in a macrokernel, and avoid nonsense like
the myriad inconsistencies in NFS.

> anyone have an opinion about why the code for printf

> is included three times in the Minix OS when it runs (once in the
> kernel, MM and FS)

Anyone have an opinion why the code for printf is included only once in
AmigaOS (even though the AmigaOS 2.04 "kernel" is actually a dozen or
more separate processes)?

Minix is a poor technology demonstrator for microkernels. Which is OK, since
it wasn't supposed to be one.

Dave Smythe

unread,

Feb 10, 1992, 2:08:22 AM2/10/92

In article <1992Feb5....@wpi.WPI.EDU> ent...@wintermute.WPI.EDU (Lawrence C. Foard) writes:
>Actually my main problem with OS theorists is that they have never tested
>there ideas! None of these ideas (with a partial exception for MACH) has ever
>seen the light of day.

David Cheriton (Prof. at Stanford, and author of the V system) said something
similar to this in a class in distributed systems. Paraphrased:

"There are two kinds of researchers: those that have implemented
something and those that have not. The latter will tell you that
there are 142 ways of doing things and that there isn't consensus
on which is best. The former will simply tell you that 141 of
them don't work."

He really rips on the OSI-philes as well, for a similar reason. The Internet
protocols are adapted only after having been in use for a period of time,
preventing things from getting standardized that will never be implementable
in a reasonable fashion. OSI adherents, on the other hand, seem intent on
standardizing everything possible, including "escapes" from the standard,
before a reasonable reference implementation exists. Consequently, you see
obsolete ideas immortalized, such as sub-byte-level data field packing,
which makes good performance difficult when your computer is drinking from
a 10+ Gbs fire-hose :-).

Just my $.02

--
========================================================================
Dave Smythe N6XLP dsm...@netcom.com (also dsm...@cs.stanford.edu)

Bill Mitchell

unread,

Feb 10, 1992, 10:03:51 AM2/10/92

in comp.os.minix, dsm...@netcom.COM (Dave Smythe) said:

>In article <1992Feb5....@wpi.WPI.EDU> ent...@wintermute.WPI.EDU (Lawrence C. Foard) writes:
>
>David Cheriton (Prof. at Stanford, and author of the V system) said something
>similar to this in a class in distributed systems. Paraphrased:
>
> "There are two kinds of researchers: those that have implemented
> something and those that have not. The latter will tell you that
> there are 142 ways of doing things and that there isn't consensus
> on which is best. The former will simply tell you that 141 of
> them don't work."
>

Yeah, but what's the odds on two who have implemented something differently
agreeing on which 141 don't work?

--
mitc...@mdd.comm.mot.com (Bill Mitchell)

Christopher Stuart

unread,

Feb 11, 1992, 8:52:57 AM2/11/92

Omniscientist

unread,

Jan 10, 2005, 9:02:10 AM1/10/05

Well, sorry Ken...but it looks like microkernels didn't last as long as
you thought. I'm on Linux right now...and minix isn't anywhere in
sight..

the...@gmail.com

unread,

Jan 11, 2005, 4:34:47 AM1/11/05

i'm glad I read this.

> >I was in the U.S. for a couple of weeks, so I haven't commented much
on
> >LINUX (not that I would have said much had I been around), but for
what
> >it is worth, I have a couple of comments now.
> >
> >As most of you know, for me MINIX is a hobby, something that I do in
the
> >evening when I get bored writing books and there are no major wars,
> >revolutions, or senate hearings being televised live on CNN. My
real
> >job is a professor and researcher in the area of operating systems.
> >
> >As a result of my occupation, I think I know a bit about where
operating
> >are going in the next decade or so. Two aspects stand out:
> >
> >1. MICROKERNEL VS MONOLITHIC SYSTEM
> > Most older operating systems are monolithic, that is, the whole
operating
> > system is a single a.out file that runs in 'kernel mode.' This
binary
> > contains the process management, memory management, file system
and the
> > rest. Examples of such systems are UNIX, MS-DOS, VMS, MVS,
OS/360,
> > MULTICS, and many more.
> >
> > The alternative is a microkernel-based system, in which most of
the OS
> > runs as separate processes, mostly outside the kernel. They
communicate
> > by message passing. The kernel's job is to handle the message
passing,
> > interrupt handling, low-level process management, and possibly
the I/O.
> > Examples of this design are the RC4000, Amoeba, Chorus, Mach, and
the
> > not-yet-released Windows/NT.
> >
> > While I could go into a long story here about the relative merits
of the
> > two designs, suffice it to say that among the people who actually
design
> > operating systems, the debate is essentially over. Microkernels
have won.
> > The only real argument for monolithic systems was performance,
and there
> > is now enough evidence showing that microkernel systems can be
just as
> > fast as monolithic systems (e.g., Rick Rashid has published
papers comparing
> > Mach 3.0 to monolithic systems) that it is now all over but the
shoutin`.
>
> Of course, there are some things that are best left to the kernel, be
it
> micro or monolithic. Like things that require playing with the
process'
> stack, e.g. signal handling. Like memory allocation. Things like
that.
>
> The microkernel design is probably a win, all in all, over a
monolithic
> design, but it depends on what you put in the kernel and what you
leave
> out.
>
> > MINIX is a microkernel-based system. The file system and memory
management
> > are separate processes, running outside the kernel. The I/O
drivers are
> > also separate processes (in the kernel, but only because the
brain-dead
> > nature of the Intel CPUs makes that difficult to do otherwise).
>
> Minix is a microkernel design, of sorts. The problem is that it
gives special
> priveleges to mm and fs, when there shouldn't be any (at least for
fs). It
> also fails to integrate most of the functionality of mm in the kernel
itself,
> and this makes things like signal handling and memory allocation
*really*
> ugly. If you did these things in the kernel itself, then signal
handling
> would be as simple as setting a virtual interrupt vector and causing
the
> signalled process to receive that interrupt (with the complication
that
> system calls might have to be terminated. Which means that a message
would
> have to be sent to every process that is servicing the process'
system call,
> if any. It's considerations like these that make the monolithic
kernel
> design appealing).
>
> The *entire* system call interface in Minix needs to be rethought.
As it
> stands right now, the file system is not just a file system, it's
also a
> system-call server. That functionality needs to be separated out in
order
> to facilitate a multiple file system architecture. Message passing
is
> probably the right way to go about making the call and waiting for
it, but
> the message should go to a system call server, not the file system
itself.
>
> In order to handle all the special caveats of the Unix API, you end
up writing
> a monolithic "kernel" even if you're using a microkernel base. You
end up
> with something called a "server", and an example is the BSD server
that runs
> under Mach.
>
> And, in any case, the message-passing in Minix needs to be completely
redone.
> As it is, it's a kludge. I've been giving this some thought, but I
haven't
> had time to do anything with what I've thought of so far. Suffice it
to say
> that the proper way to do message-passing is probably with message
ports
> (both public and private), with the various visible parts of the
operating
> system having public message ports. Chances are, that ends up being
the
> system call server only, though this will, of course, depend on the
goals
> of the design.
>
> > LINUX is
> > a monolithic style system. This is a giant step back into the
1970s.
> > That is like taking an existing, working C program and rewriting
it in
> > BASIC. To me, writing a monolithic system in 1991 is a truly
poor idea.
>
> Depends on the design criteria, as you should know. If your goal is
to
> design a Unix workalike that is relatively simple and relatively
small,
> then a monolithic design is probably the right approach for the job,
because
> unless you're designing for really backwards hardware, the problems
of
> things like interrupted system calls, memory allocation within the
kernel
> (so you don't have to statically allocate *everything* in your OS),
signal
> handling, etc. all go away (or are at least minimized) if you use a
> monolithic design. If you want the ability to bring up and take down
> file systems, add and remove device drivers, etc., all at runtime,
then
> a microkernel approach is the right solution.
>
> Frankly, I happen to like the idea of removable device drivers and
such,
> so I tend to favor the microkernel approach as a general rule.
>
> >2. PORTABILITY
> > Once upon a time there was the 4004 CPU. When it grew up it
became an
> > 8008. Then it underwent plastic surgery and became the 8080. It
begat
> > the 8086, which begat the 8088, which begat the 80286, which
begat the
> > 80386, which begat the 80486, and so on unto the N-th generation.
In
> > the meantime, RISC chips happened, and some of them are running
at over
> > 100 MIPS. Speeds of 200 MIPS and more are likely in the coming
years.
> > These things are not going to suddenly vanish. What is going to
happen
> > is that they will gradually take over from the 80x86 line. They
will
> > run old MS-DOS programs by interpreting the 80386 in software.
(I even
> > wrote my own IBM PC simulator in C, which you can get by FTP from
> > ftp.cs.vu.nl = 192.31.231.42 in dir minix/simulator.) I think
it is a
> > gross error to design an OS for any specific architecture, since
that is
> > not going to be around all that long.
>
> Again, look at the design criteria. If portability isn't an issue,
then
> why worry about it? While LINUX suffers from lack of portability,
portability
> was obviously never much of a consideration for its author, who
explicitly
> stated that it was written as an exercise in learning about the 386
> architecture.
>
> And, in any case, while MINIX is portable in the sense that most of
the code
> can be ported to other platforms, it *still* suffers from the
limitations of
> the original target machine that drove the walk down the design
decision tree.
> The message passing is a kludge because the 8088 is slow. The kernel
doesn't
> do memory allocation (thus not allowing FS and the drivers to get
away with
> using a malloc library or some such, and thus causing everyone to
have to
> statically allocate everything), probably due to some other
limitation of
> the 8088. The very idea of using "clicks" is obviously the result of
the
> segmented architecture of the 8088. The file system size is too
limited
> (theoretically fixed in 1.6, but now you have *two* file system
formats to
> contend with. If having the file system as a separate process is
such a
> big win, then why don't we have two file system servers, eh? Why
simply
> extend the existing Minix file system instead of implementing BSD's
FFS
> or some other high-performance file system? It's not that I'm greedy
> or anything... :-).
>
> > MINIX was designed to be reasonably portable, and has been ported
from the
> > Intel line to the 680x0 (Atari, Amiga, Macintosh), SPARC, and
NS32016.
> > LINUX is tied fairly closely to the 80x86. Not the way to go.
>
> All in all, I tend to agree.

>
> >Don`t get me wrong, I am not unhappy with LINUX. It will get all
the people
> >who want to turn MINIX in BSD UNIX off my back. But in all honesty,
I would
> >suggest that people who want a **MODERN** "free" OS look around for
a
> >microkernel-based, portable OS, like maybe GNU or something like
that.
>

> Yeah, right. Point me someplace where I can get a free "modern" OS
and I'll
> gladly investigate. But the GNU OS is currently vaporware, and as
far as I'm
> concerned it will be for a LOOOOONG time to come.
>
> Any other players? BSD 4.4 is a monolithic architecture, so by your
> definition it's out. Mach is free, but the BSD server isn't (AT&T
code,
> you know), and in any case, isn't the BSD server something you'd
consider
> to be a monolithic design???
>
> Really. Why do you think LINUX is as popular as it is? The answer
is
> simple, of course: because it's the *only* free Unix workalike OS in
> existence. BSD doesn't qualify (yet). Minix doesn't qualify. XINU
> isn't even in the running. GNU's OS is vaporware, and probably will
> be for a long time, so *by definition* it's not in the running. Any
> other players? I haven't heard of any...
>
> >Andy Tanenbaum (a...@cs.vu.nl)
>
> Minix is an excellent piece of work. A good starting point for
anyone who
> wants to learn about operating systems. But it needs rewriting to
make it
> truly elegant and functional. As it is, there are too many kludges
and
> hacks (e.g., the message passing).
>
> Kevin Brown

Gary

unread,

Jan 12, 2005, 9:04:45 PM1/12/05

chandan

unread,

Jan 14, 2005, 6:11:02 AM1/14/05

hi,
im new to this forum. but i can say that there is not a single
operating system which had been designed for students except minix.
ofcourse linux is great in its own terms. but doesn't u think that
minix is well suited for distributed operating system concept.i liked
microkernel concept. i certainly say that minix is student's os.
linux, too, was not great initially but the modification provided by
different groups and organization changed it into a major force. i
know that minix is older than linux but former is not popular. reason
is itself cited in one of A. Tanen. article. (want to leave this os for
students).

Adrien Plisson

unread,

Jan 16, 2005, 4:43:24 AM1/16/05

Gary wrote:
> Is there a particular reason that you're trolling this group by answering
> 13 year old posts by Ken Thompson?

i do have a question: why do i see a lot of greetings post for thorvald
on this newsgroup lately ? are these "10 years old posts revived by some
kind of bug" ? do people confuse minix and linux ?

thanks

--
rien

Adrien Plisson

unread,

Jan 16, 2005, 4:49:47 AM1/16/05

well, i didn't read the post by Michael Black when i posted this. Now i
understand, sorry for disturbing...

--
rien

Tux Wonder-Dog

unread,

Jan 20, 2005, 6:50:12 AM1/20/05

chandan wrote:

Actually, there's Switzer's Tunix, a more rigorously microkernel-based
teaching operating system than Minix - because practically everything's a
server in Tunix, versus not that many in Minix. Switzer wrote his book and
OS for students.

And there is Wirth's Oberon, just to stir the pot a bit. Wirth wrote his
experimental OS plus language for students.

And Per Brinch Hansen's Solo and associated Concurrent Pascal OSes, largely
written for students as well.

It seems to be quite wide-spread.

One of the OSes that _wasn't_ written for students, surprise, surprise, was
Unix. It was a production system that got studied - read the Lions Book/s.

Having said that, if it hadn't been for the work of Andy Tanenbaum in
writing his book and Minix, the world would've missed out on Linux, since
it inspired Linus Torvalds to go one better; and I for one wouldn't have
learned as much about Operating Systems as I've managed to.

So, thanks to Andy Tanenbaum; and also thanks to Linus Torvalds. And, did I
forget to mention, thank you to the entire BSD team through the years, etc?

Wesley Parish
--
"Good, late in to more rewarding well." "Well, you tonight. And I was
lookintelligent woman of Ming home. I trust you with a tender silence." I
get a word into my hands, a different and unbelike, probably - 'she
fortunate fat woman', wrong word. I think to me, I justupid.
Let not emacs meta-X dissociate-press write your romantic dialogs...!!!

chandan

unread,

Jan 24, 2005, 3:04:47 AM1/24/05

thanks tux. i hadn't any idea about these OSes. thanks again

burt

unread,

Jan 25, 2005, 1:36:49 PM1/25/05

Moses

unread,

May 29, 2011, 11:38:31 PM5/29/11

Moses

unread,

May 29, 2011, 11:42:16 PM5/29/11

me 3 :DDD

derrick....@gmail.com

unread,

Mar 13, 2012, 8:59:27 AM3/13/12

This thread is impressive. The idea that I can read old USENET posts is absolutely amazing. The idea I can reply to them is even cooler.
First of all, Yes, Tanenbaum was wrong, but the discussion here will last far longer than LINUX will.

Second of all, for the first internet "flame war", this was likely the most mild flame war I've ever seen. If people were "Flaming", they were certainly more cordial!

As an aside, I wonder what happens when I post here. Does it only update to Google Groups, or am I actually posting on the USENET? (I'm pretty sure it's the latter, because Groups should allow me to post to the USENET proper, but I'm not sure).

max...@googlemail.com

unread,

Mar 15, 2012, 7:19:33 AM3/15/12

Yes this is a legendary thread!

Tonton Th

unread,

Mar 20, 2012, 6:41:32 AM3/20/12

On 03/13/2012 01:59 PM, derrick....@gmail.com wrote:

>
> As an aside, I wonder what happens when I post here.
> Does it only update to Google Groups, or am I actually
> posting on the USENET? (I'm pretty sure it's the latter,
> because Groups should allow me to post to the USENET proper,
> but I'm not sure).

Welcome in the good old real Usenet.

Nous vivons dans un monde étrange/
http://foo.bar.quux.over-blog.com/

Srinivas Nayak

unread,

Jun 16, 2012, 7:28:29 AM6/16/12

One thing I heard is, linux 2.6 kernel doesn't support many old hardwares.
What is that exactly?
Is it because some device drivers removed because of bulkiness of 2.6?

tth...@panvistamedia.com

unread,

Jun 29, 2012, 5:57:33 PM6/29/12

On Tuesday, March 20, 2012 6:41:32 AM UTC-4, Tonton Th wrote:

> On 03/13/2012 01:59 PM, derrick....@gxxxx.com wrote:
>
> >
> > As an aside, I wonder what happens when I post here.
> > Does it only update to Google Groups, or am I actually
> > posting on the USENET? (I'm pretty sure it's the latter,
> > because Groups should allow me to post to the USENET proper,
> > but I'm not sure).
>
> Welcome in the good old real Usenet.

I thought Google Groups quickly fixed this mis-feature?

But with respect to the topic of the thread:

"MINIX was designed to be reasonably portable, and has been ported from the
Intel line to the 680x0 (Atari, Amiga, Macintosh), SPARC, and NS32016.
LINUX is tied fairly closely to the 80x86. Not the way to go."

How things change. As far as Minix v3 is concerned, "all the world's an x86"; downloads don't even bother naming the single supported platform: http://www.minix3.org/download/index.html

Maybe it will eventually be ported to ARM. But portability in general is, sadly, no longer a goal for the project.

mr.fi...@gmail.com

unread,

Sep 20, 2012, 9:07:18 PM9/20/12

Em quarta-feira, 29 de janeiro de 1992 11h23min33s UTC-2, ast escreveu:

> I was in the U.S. for a couple of weeks, so I haven't commented much on
> LINUX (not that I would have said much had I been around), but for what
> it is worth, I have a couple of comments now.
>

> nature of the Intel CPUs makes that difficult to do otherwise). LINUX is

> a monolithic style system. This is a giant step back into the 1970s.
> That is like taking an existing, working C program and rewriting it in
> BASIC. To me, writing a monolithic system in 1991 is a truly poor idea.
>
>

> 2. PORTABILITY
> Once upon a time there was the 4004 CPU. When it grew up it became an
> 8008. Then it underwent plastic surgery and became the 8080. It begat
> the 8086, which begat the 8088, which begat the 80286, which begat the
> 80386, which begat the 80486, and so on unto the N-th generation. In
> the meantime, RISC chips happened, and some of them are running at over
> 100 MIPS. Speeds of 200 MIPS and more are likely in the coming years.
> These things are not going to suddenly vanish. What is going to happen
> is that they will gradually take over from the 80x86 line. They will
> run old MS-DOS programs by interpreting the 80386 in software. (I even
> wrote my own IBM PC simulator in C, which you can get by FTP from
> ftp.cs.vu.nl = 192.31.231.42 in dir minix/simulator.) I think it is a
> gross error to design an OS for any specific architecture, since that is
> not going to be around all that long.
>

> Don`t get me wrong, I am not unhappy with LINUX. It will get all the people
> who want to turn MINIX in BSD UNIX off my back. But in all honesty, I would
> suggest that people who want a **MODERN** "free" OS look around for a
> microkernel-based, portable OS, like maybe GNU or something like that.
>
>

> Andy Tanenbaum (a...@cs.vu.nl)
>
>
> P.S. Just as a random aside, Amoeba has a UNIX emulator (running in user
> space), but it is far from complete. If there are any people who would
> like to work on that, please let me know. To run Amoeba you need a few 386s,
> one of which needs 16M, and all of which need the WD Ethernet card.

Yeah, this is the famous and epic discussion between Linus Torvalds and Andrew S. Tanembaum about Linux kernel architecture and how Tanembaum thinks that monolithic kernels are inferior to microkernels.

mkam...@tvz.hr

unread,

Sep 28, 2012, 9:45:52 AM9/28/12

Dana srijeda, 29. siječnja 1992. 14:23:33 UTC+1, korisnik ast napisao je:

Wow,wish that either Linus or Tannenbaum would be my teachers...If any of you two ever start to work for the Technical Polytechnic of Zagreb I will be very happy.

nimeto...@gmail.com

unread,

Oct 9, 2012, 2:57:07 PM10/9/12

This shoulda been locked.

Martijn van Buul

unread,

Oct 10, 2012, 8:07:26 AM10/10/12

delro...@gmail.com

unread,

Oct 22, 2012, 8:35:08 PM10/22/12

to Martijn van Buul

Martijn van Buul

unread,

Oct 23, 2012, 2:51:48 AM10/23/12

* delro...@gmail.com:

Maybe I'm a zebra. That, or maybe you have no clue what you're talking about.

I see it's obviously the latter explanation.

Edward A. Falk

unread,

Oct 23, 2012, 8:37:48 PM10/23/12

Asad Dhamani

unread,

Nov 18, 2012, 3:33:29 PM11/18/12

Welcome to 2012! 20 years, I have my own Linux distro! :D The future is now a thing of the past! Feels so great seeing so old threads!

amyas...@gmail.com

unread,

Nov 25, 2012, 6:30:50 AM11/25/12

So I can post a reply right now for this historical flame war?? And Google will actually index my name along with ast and Linus? Wow :D

luke...@gmail.com

unread,

Nov 26, 2012, 2:34:02 PM11/26/12

El miércoles, 29 de enero de 1992 09:20:50 UTC-5, David Megginson escribió:
> I would like to at least look at LINUX, but I cannot, since I run
> a 68000-based machine. In any case, it is nice having the kernel
> independent, since patches like the multi-threaded FS patch don't
> have to exist in a different version for each CPU.
>
> I second everything AST said, except that I would like to see
> the kernel _more_ independent from everything else. Why does the
> Intel architecture _not_ allow drivers to be independent programs?
>
> I also don't like the fact that the kernel, mm and fs share the
> same configuration files. Since they _are_ independent, they should
> have more of a sense of independence.
>
>
> David
>
> #################################################################
> David Megginson meg...@epas.utoronto.ca
> Centre for Medieval Studies da...@doe.utoronto.ca
> University of Toronto 39 Queen's Park Cr. E.
> #################################################################

ast, time proved you were wrong, yet, you are the best dinosaur I ever know,
greetings from cuba.

Skywalker.

Martijn van Buul

unread,

Nov 27, 2012, 5:33:38 AM11/27/12

caront...@hotmail.com

unread,

Dec 30, 2012, 5:32:31 PM12/30/12

El miércoles, 29 de enero de 1992 06:23:33 UTC-7, ast escribió:

sk...@ns.sympatico.ca

unread,

Feb 1, 2013, 6:14:26 AM2/1/13

to Martijn van Buul

On Wednesday, October 10, 2012 9:07:26 AM UTC-3, Martijn van Buul wrote:

> > This shoulda been locked.
>
>
>
> This isn't a forum.
>

it's usenet, afaik a non-moderated newsgroup can't really have locked threads, violates the rules.

Martijn van Buul

unread,

Feb 1, 2013, 6:53:06 AM2/1/13

* sk...@ns.sympatico.ca:

> it's usenet, afaik a non-moderated newsgroup can't really have locked
> threads, violates the rules.

It's not a question of "rules", it's a question of not being capable of.

The rules, if any, state you shouldn't post articles with a line length of
more than 80 characters ;)

derrick....@gmail.com

unread,

Apr 17, 2013, 12:31:22 PM4/17/13

I feel like I'm a part of history.
I wasn't even born when this thread opened...

syn4...@gmail.com

unread,

Apr 25, 2013, 1🔞37 PM4/25/13

Me too. :D

usama...@gmail.com

unread,

Apr 28, 2013, 4:57:13 AM4/28/13

I am very surprised to read this thread.
Was there Google and Google groups in 1992.....?

Antoine Leca

unread,

Apr 29, 2013, 6:58:44 AM4/29/13

usama...@gmail.com wrote:
> I am very surprised to read this thread.
> Was there Google and Google groups in 1992.....?

Not exactly; there was known as Usenet then; and Google Groups just
bought the Usenet archives (through DejaNews) in 2001 IIRC.
http://www.google.com/googlegroups/archive_announce_20.html

And since Usenet does allow threads to last and last, and since Pr
Tanenbaum did not put an "Expire" header on his post, Google Groups
offer you the opportunity to post to that thread. Even posts completely
off-topic to that thread (or to this group comp.os.minix as well.)

Antoine

Michael Black

unread,

Apr 30, 2013, 3:35:44 PM4/30/13

No.

Google moved to a new interface, and allowed replies to messages older
than 30 days. 30 days wsa the way from the beginning, and really is
proper for old messages. They had it that way from when they took over
the dejanews archive. A previous time they changed the interface, and put
the bug in, but after complaints it was fixed. I have no idea if they
removed the vandalism to old threads, but that time back some years did
cause stupid responses just like now, the idiots attracted to a thread
that google itself has pointed out as "historic", at one point they put up
a timeline of usenet and pointed to specific posts, including this one.

I'm tired of fighting with google, so I've never bothered complaining this
recent tiem. Their current interface may be fine for a web-based
newsgroup, but it's not right at all for Usenet. It doesn't even show
dates, or even where the message is posted to (so nobody knows about
cross-posting.

There is no reason to reply to old messages. The conversation has moved
on, if someone saved the message at the time and gets around to it later,
that's different from someone replhing through google years later. The
posters just drop the thread into the newsgroup, they are oblivious to
where it's going. They often don't even quote. Suddenly a mystery
message appears in the newsgroup without context, people reply without
even wondering where the rest of the thread is.

The replies usually offer nothing new, they ignore that the thread back 20
years ago had plenty of answers, likely sufficient. The replies ignore
the fact that whoever posted back when may no longer be reading the
newsgroup. Indeed, when some idiot replies to an old message that was
offereing something for sale, the original poster may not have been there
except to post his ad.

All of these replies are from idiots who think it's "cool" to reply to an
old thread, like they were actually around 20 years ago. They haven't
even added anything useful, just a bunch of "me toos".

The only reason I've not bothered replying to condemn these idiots is
because this is a historical thread, and I didn't want to add my
vandalism. But since you've just given approval to it, I had to speak
out. Don't ecnourage the idiots.

As an aside, deajnews only started in 1995 or 96. Their archive is what
google bought. But then google tracked down older archives of a more
limited nature, putting them together, which is why there is now an
incomplete archive going back to the start of Usenet, 1979. It's those
other archives that kept this thread, and the vandalism is because of
google, and because of the idiots who reply.

Michael