[TUHS] Mach for i386 / Mt Xinu or other

Larry McVoy lm at mcvoy.com
Wed Feb 22 02:47:28 AEST 2017


On Tue, Feb 21, 2017 at 07:02:18AM -0500, Noel Chiappa wrote:
> So there is a question here, though, and I'm curious to see what others who
> were closer to the action think. Why _did_ Linux succeed, and not a Unix
> derivative? (Is there any work which looks at this question? Some Linux
> history? If not, there should be.)

http://www.mcvoy.com/lm/bitmover/lm/papers/srcos.html

is worth a read, I was very much in the middle of all this at the time.

I think Linux succeeded because:

    - it was free and GPLed.  BSD license is nice but it has the problem
      that people can take it closed source and not give back changes.

    - no lawsuit

    - Linus (as mentioned, much stronger leader than any in the Unix world)

    - no religion.  I can't make this point hard enough.  At Sun, we couldn't
      change any API, any utility, it was compat to the point that it was not
      useful.  Look at SVr4/Solaris /proc and then look at Linux /proc.  The
      Linux one is way, way, way, way more useful, you can dig shit out with
      shell scripts.  The rest of the Unix world was blindly posix compat
      even when posix compat made no sense.  Linux was glorious in that 
      Linus wanted compat but was willing to break it for good reasons.

    - good enough

    - fun, all the cool kids were there.

Here is perhaps something that will resonate with this crowd.  When I was
teaching OS at Stanford, for file systems I made people go through the
thought process on when you write what (think the sync writes so the FS
isn't busted when you crash).

For years, the BSD guys insisted that Linux was doing it wrong because
they didn't do the sync writes, they did ordered writes (but the BSD
guys didn't understand the write ordering so they thought it was busted,
as did I).  

But Linux wasn't busted, they had carefully gone through the process
of figuring out when stuff had to be first so that you could survive
a crash.

The BSD guys refused to believe that it was possible, they were stuck
on writes had to be synchronous in order to get a file system that 
wasn't corrupted on a crash.

I eventually started convincing them that Linux had it right by saying
just untar some big tarball and unplug the power in the middle of that
on a system with UFS and a system with ext2.  Tell me what happens when
you power it up.  Answer?  The UFS based system dropped you into a fsck
nightmare of unattached inodes, the ext2 based systems lost some data
but the file system was fine.

Obviously, the Linux approach won, it's better.  Way better, higher 
performance and a much better user experience.  But the traditional
Unix guys had to be dragged kicking and screaming, over a decade, 
into that world.  It's stuff like that that made Unix stagnate while
Linux forged ahead.


More information about the TUHS mailing list