Subject: Re: LFS (was Thank you NetBSD)
To: Jochen Kunz <jkunz@unixag-kl.fh-kl.de>
From: Thor Lancelot Simon <tls@rek.tjls.com>
List: tech-kern
Date: 02/23/2005 14:26:03
On Wed, Feb 23, 2005 at 08:15:09PM +0100, Jochen Kunz wrote:
> 
> I have to admit that I have nearly no knowledge on file system internas.
[...]
> NetBSD. AdvFS has _really_ advanced concepts and features compared with
> any other FS knowen to me. At least somthing like XFS would be nice.

I'm sorry, I see more than a slight contradiction here.  Is it just me?

There are plenty of paths not taken that would be better than the one
most operating systems take to get from point A to point B.  Often the
path you're on is a result of who implemented what, with less bugs, when,
and happened to have the ear of enough other people to get it shipped
in volume (so the bugs that were there got fixed), not actually a result
of it being the shortest or straightest path to where you want to go.

That's true for asynchronous I/O, for cache replacement, for the
internals of the network stack, even for data structures.  Look how
many years passed between the first running kernel whose data structures
were designed for lockless concurrent access (Synthesis) and the current
rage for that technique, for example (perhaps direct code overlay will
be next, though somehow I doubt it ;-)).

The basic problem with the BSD LFS is that the quality of the
implementation is so low that it's almost not usable.  That says
precisely nothing about the utility of LFS as a general structure for
a filesystem; indeed, the Sprite papers are as persuasive now as they
were a decade ago, and those folks had _real_ results from code that
actually _worked_ (I know, I used it).

I tend to count some of the egregious design changes made in BSD LFS
as "low quality implementation".  I suppose we could argue about that,
but really it's just splitting hairs.  If you want to see an example
of an LFS-like filesystem that's been wildly successful in the real
world, WAFL is a pretty good one.

-- 
 Thor Lancelot Simon	                                      tls@rek.tjls.com

"The inconsistency is startling, though admittedly, if consistency is to be
 abandoned or transcended, there is no problem."		- Noam Chomsky