Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 26 Apr 1997 13:10:33 -0700 (MST)
From:      Terry Lambert <terry@lambert.org>
To:        joa@kuebart.stuttgart.netsurf.de (Joachim Kuebart)
Cc:        sysop@mixcom.com, hackers@freebsd.org
Subject:   Re: VFAT 32 support in msdosfs
Message-ID:  <199704262010.NAA07503@phaeton.artisoft.com>
In-Reply-To: <XFMail.970426112148.joa@delos.lf.net> from "Joachim Kuebart" at Apr 26, 97 10:33:11 am

next in thread | previous in thread | raw e-mail | index | archive | help
> >Much rather have NTFS in 95 with all the little BS files less then 1K.
> 
> Does NTFS have owners and permissions like any file system should? Besides, I
> don't understand why they didn't just use NTFS... Only so now there's 2
> _incompatible_ bad file systems ?

Despite the claims to the contrary, the only real win in NTFS is the
directory btree's, and then only for what a UNIX user would consider
a *really* full directory.  In general, NTFS performs worse than VFAT.

With VFAT hogging all free memory as much as it can, and with the
full directory (mostly) in core (see the MACH paper, which turns off
all the performance attributes of FFS as "unfair" and then loads the
entire FAT into core as "a necessary optiomization"), you would be
hard pressed to beat VFAT in a race.  We were able to do it with
FFS in the Heidemann framework under Windows 95, and then only got
really serious performance improvement after implementing our own
paging handler and soft updates.  We had some really clever people.

Log structuring, and all the associated BS, is only really useful if
you have database transactions, and even then only if the transaction
interface is exported to the top level.  The performance and reliability
comparisons to FFS are all pretty much bogus.  With a coulpe of minor
tweaks (like setting up the default block size from a measly 4k), it
was possible to get the UFS (FFS) in UnixWare to perform withing 5%
of VXFS on the same hardware.  So much for the touted improvements,
other than the transactioning features.

The only people to really take off with this type of FS design are
Network Appliance Corp., the people who got Guy Harris after Auspex.
They get their winds from FFS-style geometry optimization, using
stripe sets ias if they were physical geometry artifacts.  Quite
clever, actually, but not terribly suitable for paging access, as
in interactive use, as oppossed to be a server on a wire (it's to
be expected: it was *designed* to be a server, so that's the right
trade).


If you are really interested in NTFS, I recommend:

	The read-only NTFS code available for Linux


	Inside the Wondows NT File System
	Helen Custer
	Microsoft Press
	ISBN 1-55615-660-X


	Inside Windows NT
	Helen Custer
	Microsoft Press
	ISBN 1-55615-481-X

The first goes into enough detail that you could write a read-only
NTFS, if you had one to test your code on.

Get the second only if you are interested in the NT security model,
since it has little real information on the FS itself, other than
explaining why some of the FS stuff from the first book is the way
it is.


					Regards,
					Terry Lambert
					terry@lambert.org
---
Any opinions in this posting are my own and not those of my present
or previous employers.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199704262010.NAA07503>