[RFC] Remove NTFS kernel support

[Szabolcs Szakacsits]

John-Mark Gurney <jmg <at> funkthat.com> writes:
> Nikolay Pavlov wrote this message on Thu, Feb 07, 2008 at 20:52 +0200:
> > On Thursday 07 February 2008 20:18:42 Joao Barros wrote:
> > > On Feb 7, 2008 5:41 PM, Nikolay Pavlov <qpadla <at> gmail.com> wrote:
> > > > On Thursday 07 February 2008 14:47:41 Eric Anderson wrote:
> > > > > FUSE is slow, requires a port (unless PUFFS is ported, which I've
> > > > > probed about before).
> > > >
> > > > I think this is not an argument:
> > > > http://www.ntfs-3g.org/performance.html
> > >
> > > Eric has valid points.
> > > How relevant is a benchmark on Linux to your argument?
> > 
> > But it's a userland application. This page is demonstration of it's 
> > potential performance that could be achieved, but were is the FreeBSD NTFS 
> > implementation stats? Let me ask you: compered to what FUSE is slow?     
> 
> Kernel NTFS support is about 10x faster than ntfs-3g on FreeBSD (I
> think this also depends upon the size of the file).  This is because
> ntfs-3g depends upon the block device that linux provides to userland.
> There are patches that make ntfs-3g have it's own block cache that
> makes it perform decently on FreeBSD, but until those patches are
> integrated, using ntfs-3g is a non-starter if you use NTFS for >4GB
> file support.  It's faster to use samba to a Windows box than it is
> to use ntfs-3g to write large files.  (And that's even w/ how much
> slower samba is that nfs.)
> 
> I don't have any hard core benchmarks handy.  Even on MacOSX ntfs-3g
> is sooo slow.  It's so slow, that I don't even both hooking up NTFS
> disks to my MacOSX box anymore either.
> 
> Though I will say that once ntfs-3g has decided that they want to
> target other platforms than Linux and address these performance issues,
> I will be one of the first asking for our current NTFS code to be
> removed and replaced by ntfs-3g, but until that time, we need to keep
> the current code.

Well, I think we have worked hard all last year with Csaba Henk 
(FUSE/FreeBSD and the ublio caching layer author/maintainer), 
Alejandro Pulver (fusefs-ntfs, ntfs-3g/FreeBSD developer/maintainer),
Paul Marks (ntfs-3g/OS X ublio integrator/ex-maintainer) and
Erik Larsson (ntfs-3g/OS X ublio integrator/developer/maintainer) 
to provide this performance improvement by default on FreeBSD and 
OS X. 

The FreeBSD one is available for seven months:

 http://www.freshports.org/sysutils/fusefs-ntfs

I think this is very worth to read:

http://www.freebsd.org/cgi/cvsweb.cgi/ports/sysutils/fusefs-ntfs/files/pkg-message.in?rev=1.6

The OS X one is here for four months:

 http://macntfs-3g.blogspot.com/

The UBLIO DMG is this one:
 
 http://hem.bredband.net/unsound/ntfs-3g/NTFS-3G_1.2129-ublio-catacombae.dmg

The most common reasons for performance issues (besides the driver being
unoptimized on all platforms) are listed here:

 http://ntfs-3g.org/support.html#cpu100

Extremely slow external USB performance is/was very typical 
because of this issue (USB2->USB1) on several OSes:

 http://www.theinquirer.net/gb/inquirer/news/2007/12/06/seagate-snubs-linux

Regards,   Szaka