Hello.

I have just sent this e-mail to the [EMAIL PROTECTED] mailing
list, I hope this information is of any help to you.

The mentioned patch for the FreeBSD port is here:
http://people.freebsd.org/~alepulver/fusefs-ntfs.diff
Or, as a .tar.gz:
http://cvsweb.freebsd.org/ports/sysutils/fusefs-ntfs/fusefs-ntfs.tar.gz?tarball=1

And the rest here:
http://www.freebsd.org/cgi/cvsweb.cgi/ports/sysutils/fusefs-ntfs/

The original message is below:
--------------------

Hello.

I have applied the libublio patch to the current ntfs-3g and
integrated
with fjoe's aligned I/O layer. Before the read/write speed was about
2/1.2 MB/s, and now about 15/9 MB/s (of course depends on many things,
but to illustrate the point that is drastically improves performance).
The ni-kernel driver reads at 10MB/s, but is much faster in filesystem
operations (listing files, find(1), etc). In addition the disk
activity
is a lot smoother. Also it outputs an error when creating a file
sometimes, but can be safely ignored (just try again), I will
investigate it (this happened before too).

The reason AFAIK is the lack of cache for block devices, which was
(re)added in FreeBSD-CURRENT (7.x). So the solution for users of
FreeBSD 6 (and 5, but the port isn't available for that release)
consists on using a user space cache.

I tried to contact fuse4bsd/libublio author, port maintainer, and
asked
in freebsd-hackers@/freebsd-performance@ without response (I am not
blaming them, perhaps they were busy, I didn't ask correctly or simply
there was no interest about this there). I don't even know what's the
OSVERSION value to check this.

That's why I am asking for testers here. The patch to the current port
is attached, if libublio-20070103.tar.gz can't be fetched because the
mirrors weren't updated get it from:
http://people.freebsd.org/~alepulver/).

Without UBLIO_SYNC_IO=0 it's slower than without UBLIO, and increasing
UBLIO_BLOCKSIZE to 1/2/4MB seems a good improvement (also the disk is
not constantly reading/writing). Increasing UBLIO_ITEMS more than 24
doesn't seem to increase performance.

In Mac OS X the same problem exists, but cache for block devices was
not added AFAIK so this would be interesting (currently the only
working solution) for them.

Best Regards,
Ale


--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups 
"macfuse-devel" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at 
http://groups.google.com/group/macfuse-devel?hl=en
-~----------~----~----~----~------~----~------~--~---

Reply via email to