cvs commit: src/sys/fs/ntfs ntfs_subr.c

[Jeremie Le Hen]

Hi Robert,

On Thu, Apr 03, 2008 at 10:48:03AM +0100, Robert Watson wrote:
>  On Thu, 3 Apr 2008, Jeremie Le Hen wrote:
> > For the records, Ithink NetBSD has developped a BSDish solution 
> > API-compatible with FUSE, ironically named refuse :-).  This is just a 
> > compatibility layer on top of a kernel based component named puffs.
> >
> > http://www.netbsd.org/docs/puffs/
> >
> > Maybe this should end up in the project idea page.
> 
>  We currently have two user file system frameworks in ports (fuse, Arla 
>  nnpfs) and one in the kernel already (Coda).  And no doubt more I don't know 
>  about. We're actively discussing importing nnpfs to make access to AFS 
>  easier out of the box.  I'd be careful to recommend a determination of 
>  whether it makes a useful further contribution or not, rather than just 
>  recommending a port of the code.  I have a loose recollection from the 
>  AsiaBSDCon puffs talk last year that fast path access to local cache 
>  (container) files wasn't implemented in puffs, which would put it at a 
>  significant disadvantage performance-wise to nnpfs.  However, a year is a 
>  long time, so it may well be that puffs has now picked up that support in 
>  the mean time.

I don't know anything about Arla or Coda, so I won't make any
performance-wise comments about those filesystem.

However, there seem to be quite a community around FUSE.  I think this
is mostly because people are more confident with userland coding.  There
are really plenty of filesystems using FUSE, albeit most of them seem to
be very silly IMHO.  See http://fuse.sf.net/wiki/index.php/FileSystems .

Yet some of them could be useful for people needing compatibility before
performance.  NTFS is a good example, FUSEPod, SshFS are good ones too.
Once we'd have gathered FreeBSD fellows around puffs/refuse, some of
them may tackle the performance problem.

Regards,
-- 
Jeremie Le Hen
< jeremie at le-hen dot org >< ttz at chchile dot org >