Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 27 May 1998 09:49:56 +0200
From:      Martin Cracauer <cracauer@cons.org>
To:        Amancio Hasty <hasty@rah.star-gate.com>, Martin Cracauer <cracauer@cons.org>
Cc:        Ted Stein <ted@taki.net>, advocacy@FreeBSD.ORG
Subject:   Re: Braindead FreeBSD advocates (Re: Advocacy Shortcomings)
Message-ID:  <19980527094956.00484@cons.org>
In-Reply-To: <199805261556.IAA04155@rah.star-gate.com>; from Amancio Hasty on Tue, May 26, 1998 at 08:56:39AM -0700
References:  <19980526121304.51425@cons.org> <199805261556.IAA04155@rah.star-gate.com>

next in thread | previous in thread | raw e-mail | index | archive | help
In <199805261556.IAA04155@rah.star-gate.com>, Amancio Hasty wrote: 
> The debate at slasdot.org was fine . The FreeBSD advocacy was not
> brilliant however the linux advocay was even worse.

Well, from a certain standpoint the FreeBSD votes may have been
reasonable in that their fought the Linux folks with their own
weapons (at least the weapons they use in non-linux dominated
forums). 

However, I think the FreeBSD folks have to do better as they fight
from an inferior position. Unreasonable FreeBSD advocacy hurts more
than unreasonable Linux advocacy.

> The next time an article gets published at slashdot.org just try 
> to follow up with a few good posts.

Tried to do so.
 
> With respect to advocates, basically, the group needs to come up with a 
> few good themes or stories which can be backed up with data.

One thing I'd really like is a down-to-metal explanation of the
(a)sync metadata issue. Describe exactly under what circumstances the
typical Linux filesystem is in a state that will loose files or the
whole filesystem.

Theo De Raadt came up with a good example: A simple rename(2)/mv(1) is
usualy thought to be an operation that can't lose the file. Not so on
Linux. 

Explanation in my (half-baken) words: Since metadata is async and
writes are completely unsorted, about half of the rename operations
have an intermediate state where the old filename is deleted but the
new one isn't in place. Since Linux also doesn't have any contraints
about the chunks written out, allocatition and connection of the new
name may also happen an arbitrary long time after loosing the old
name. If your system crashes within that time, you still have the data
blocks and the indoe, but you'll the data unnamed in lost+found. The
latter also concerns the quality of Linux fsck, which I'm not sure
about, maybe the file gets lost completely.

I'm not sure I understand enough of the issue to make a bullet-proof
webpage with some diagrams. This requires reading quite a lot of Linux
kernel/fsck/mkext2fs code. For FreeBSD, I also noted that the comments
in the 4.4BSD book about metadata synchronisation don't meet reality
in FreeBSD.

Martin
-- 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Martin Cracauer <cracauer@cons.org> http://www.cons.org/cracauer
BSD User Group Hamburg, Germany     http://www.bsdhh.org/

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-advocacy" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19980527094956.00484>