FreeBSD's problems as seen by the BSDForen.de community

Andrew Reilly andrew-freebsd at areilly.bpc-users.org
Fri Jan 11 22:08:21 PST 2008


I really appologise for adding to this thread, but I'm going to
anyway.

On Fri, 11 Jan 2008 15:03:01 +0100
Timo Schoeler <timo.schoeler at riscworks.net> wrote:

> Second Bingo. And I second that it gets better again. The 5.x was a
> deep valley to walk through, at least for me.
> 
> > Peter Holm's stress testing suite, Kris
> > Kennaway's constant hammering of FreeBSD to breaking point and
> > thorough investigation of failure mode, and countless new individuals
> > like myself working in their own small way to improve the OS they
> > care for have, in my opinion, produced in FreeBSD 7 (and partially in
> > later 6.x releases), performance and stability that hasn't been seen
> > since FreeBSD 4.    
> 
> That sounds like the future I want to see. In general, I'm almost an
> ascetic in those things, but it's enough to make me smile. Thanks.

Most of the people who pine for the stability of FreeBSD-4
seem to be missing the point that it was stable essentially
because it was the well-honed linear decendent of thirty years
of single-processor Unix kernel development.  That was wonderful
and all, but it was clear by then that that was a dead-end
that was untennable to stay in: multiprocessors (including
single-chip multi-core) were coming, and it was important that
FreeBSD be there to support them.  Solaris and Linux might look
more stable on multi-processors now, but that's mostly because
they started the difficult conversion process earlier, and
so have come further down that path.  In my opinion, SunOS-5
(aka Solaris-1) was as much a step backwards from SunOS-4 as
FreeBSD-5 was from FreeBSD-4.  That's because you can make a
lot of simplifications, assumptions and shortcuts when there's
only one thread of execution.  Those assumptions are all there
in the code-base, and they're all wrong, as the code comes
out from under the GIANT lock.  Stabilization is the process
of reverse-engineering those assumptions (or replacing whole
chunks of functionality with new code that doesn't have them),
and it's very, very difficult.  Had to happen, though, and
everyone's happier now that it did.  FreeBSD-7 is by far the
best version of FreeBSD that I've used, even though it has to
support more types of processor, in more configurations, and
with more sophisticated busses and peripherals that are harder
to reverse-engineer or acquire documentation for than ever.
It's a truly amazing piece of work.  I'm looking forward to -8
keenly, but I'm happy for the developers to take as long about it
as they need.

Cheers,

-- 
Andrew


More information about the freebsd-current mailing list