Article 5041 of alt.sys.pdp10:
Path: news3.best.com!news2.best.com!newsfeed.berkeley.edu!newsfeed.enteract.com!news.enteract.com!not-for-mail
From: David Scheidt <dscheidt@enteract.com>
Newsgroups: alt.folklore.computers,alt.sys.pdp10
Subject: Re: How in Hell did the Great Unix to NT Migration begin??
Date: 6 May 1999 02:52:37 GMT
Organization: EnterAct L.L.C. Turbo-Elite News Server
Lines: 42
Message-ID: <7gr05l$6o1$1@eve.enteract.com>
References: <371be3e4.0@newsfeed.one.net> <7gn06k$7hn$1@eve.enteract.com> <7gqle7$h6j$1@shell3.ba.best.com> <7gqmbp$r3s$1@eve.enteract.com> <7gqtu5$s3j@weyl.math.psu.edu>
NNTP-Posting-Host: 207.229.143.42
User-Agent: tin/pre-1.4-19990216 ("Styrofoam") (UNIX) (FreeBSD/3.1-STABLE (i386))
Xref: news3.best.com alt.folklore.computers:129606 alt.sys.pdp10:5041

In alt.folklore.computers Alexander Viro <viro@weyl.math.psu.edu> wrote:
: In article <7gqmbp$r3s$1@eve.enteract.com>,
: David Scheidt  <dscheidt@enteract.com> wrote:
:>
:>No.  You clearly missed mine.  I have never needed an ext2ed like utility on 
:>a FFS-like filesystem.  

: Then do your friggin' homework and find the papers describing the design of
: ext2. It is derived from FFS. Or do the same homework by getting *any*
: *BSD and comparing UFS and EXT2 drivers. Ferchrissake, they share 3/4 of
: code. Differences between EXT2 and UFS being:
: 	a) EXT2 doesn't use fragments
: 	b) EXT2 keeps names in directory entries without '\0'
: 	c) EXT2 has slightly different order of fields in on-disk inodes
: and superblock.

I didn't think EXT2 used cylinder groups, either.
`
: Sheesh... If you want to start old sync vs. async flamewar - at least take
: care to RTFDejaNews and learn WTF it is about.

The syncronous metadata writing are a very important part of FFS stability.[1]
I have spent much too much time fixing ext2fs filesystems that have died 
while doing huge amounts of file creation and deletion to be convinced 
otherwise.  I have never seen the behavior on FFS FSes mounted sync, and have 
with ext2fs, and ffs mounted async.

David

[1] Gregory R. Ganger and Yale N. Patt wrote a paper a couple of years ago
which explained that you needn't actually do the metadata syncronously, but
merely in dependent order.  Their paper is available at 
http://www.ece.cmu.edu/~ganger/papers/CSE-TR-254-95/.  An implementation of 
this by Kirk McKusick is available in FreeBSD 3.0 and later (and probably 
elsewhere).  I haven't had any spectacular system failures since I started 
to use it, haven't heard of many problems with it. 

-- 
David Scheidt <dscheidt@enteract.com>
Large fibreglass fruits are much the same the world over.
                                                    -- Vicki Parslow Stafford