DragonFly kernel List (threaded) for 2003-08
Re: High Performance File Systems
> Define high-performance?
> Personally, after having had my share of VxFS, JFS, etc, etc on
> Solaris, HP-UX, Tru64, AIX, and god knows what other platforms,
> the only file system which I was left with respect for was AdvFS
> from DEC.
> Hiten Pandya has done a lot of porting work on XFS for all I know.
Yes, I did work on JFS originally for porting it to FreeBSD.
I took interest in XFS when I heard that Alexander Kabaev was
doing some work on it.
IIRC, the problems encountered in doing port of XFS to FreeBSD
was serialisation problems in the buf system. Apart from that,
he (Alexander) had almost everything sussed out; but don't take
my work for it as this was quite a WHILE ago. 8-)
On the JFS front; it's what inspired me to write our UUID
library, so that we can assign UUIDs to all mounted file systems
thus giving us ability to unmount a device by (registered) UUID
and not using the /dev/blah path system. It can be helpful in
some situations. Take Ian Dowse's work on umounting by fids.
I had jfsutils fully working, but lack of support from people
made me stop working on JFS kernel-side support. IMHO, Linux
has so many file systems, and people seem to be saying:
"Oh.. it would be great to have Filesystem X and Y and Z
So I think the goal should be focused in offering a kernel
library which offers as much ease of porting Linux VFS modules
as possible. The only difficult part of this would be providing
support for the various Linux VM/Page cache stuff.
So for more than two reasons, I think Linux file systems need
to be tackled in a different way than doing traditional porting
because a) it's too tedious (re-inventions) b) hard to keep up
in sync with main code.
I currently lack hardware to work on any file system or related
projects, otherwise I wouldn't mind working on something like a
"lib-linuxvfs" 8-) 8-)
FWIW, the mostly simplest file systems to port from Linux are
FreeVxFS, and QNX-FS.
Ok... CALL Me crazy, but... ;-)