Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 2 Mar 1998 11:00:26 -0500 (EST)
From:      "John S. Dyson" <dyson@FreeBSD.ORG>
To:        jonny@coppe.ufrj.br (Joao Carlos Mendes Luis)
Cc:        eivind@yes.no, paulg@interlog.com, roberto.nunnari@agie.ch, freebsd-chat@hub.freebsd.org
Subject:   Re: FreeBSD vs. Linux ?
Message-ID:  <199803021600.LAA13646@dyson.iquest.net>
In-Reply-To: <199803021540.MAA05086@gaia.coppe.ufrj.br> from Joao Carlos Mendes Luis at "Mar 2, 98 12:40:00 pm"

next in thread | previous in thread | raw e-mail | index | archive | help
Joao Carlos Mendes Luis said:
> #define quoting(Eivind Eklund)
> // FreeBSD is presently in the process of integrating Kirk McKusick's
> // Soft Updates code, which will give the same (actually, sometimes
> // better) performance without removing safe recovery.
> 
> Is there an URL with more info on osft updates ?  What is it ?  How
> will it affect performance, etc. ?
> 
Much of the time, when people might say "Linux is faster than FreeBSD",
it is likely a manifestation of the conservative filesystem metadata
update policy of FreeBSD.  This is (and don't let nay-sayers try to
convince you otherwise) a carefully considered and wise policy on the
part of the original BSD development group.

There is an innovative mechanism as described in a research report by
Ganger and Patt (sp???).  Recently, Kirk McKusick did a full implementation
of such, with the benefit of a consistant filesystem without the conservative
(slow) metadata policy.  This code that Kirk wrote was over 100K of source
to implement this metadata update policy, and frankly is faster than our
async (unsafe) metadata policy, and nearly as safe (modulo data loss), as
our conservative policy.

With any async write policy, you can/will loose data upon power loss or
system failure.  The problem with async write policies beyond that data loss
is the destruction of meta-data on disk, and the possibility of filesystem
damage that can't be correctly repaired by fsck.  The filesystem might look
repaired, but really isn't.  This new (soft update) mechanism keeps the
filesystem metadata safe (and is provably repairable, unless there is
a severe hardware malfunction), while giving the performance of async
mounts (actually is better.)

This thing IS an innovation, and was implemented by Dr. Kirk McKusick,
who originally wrote/designed FFS, upon which alot of the U**X filesystem
work has been based (including EXT2FS.)

-- 
John                  | Never try to teach a pig to sing,
dyson@freebsd.org     | it just makes you look stupid,
jdyson@nc.com         | and it irritates the pig.

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



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199803021600.LAA13646>