Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 11 Jun 2011 23:58:39 -0400
From:      David Magda <dmagda@ee.ryerson.ca>
To:        Volodymyr Kostyrko <c.kworr@gmail.com>
Cc:        freebsd-fs@freebsd.org, freebsd-stable@freebsd.org, Martin Matuska <mm@freebsd.org>
Subject:   Re: HEADS UP: ZFS v28 merged to 8-STABLE
Message-ID:  <525D503A-240C-49F2-9AAD-EC8E3C1CDB9A@ee.ryerson.ca>
In-Reply-To: <4DF28BCF.3060008@gmail.com>
References:  <4DECB197.8020102@FreeBSD.org> <4DF28BCF.3060008@gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Jun 10, 2011, at 17:25, Volodymyr Kostyrko wrote:

> Am I missing something? How about using fletcher[24] for dedup?

Fletcher is fairly weak as things go, and so even though two checksums =
are the same, there's a decent chance that the data is actually =
different. At least with recent releases of (Open)Solaris, when you =
enable do a 'dedup=3Don' the has used is SHA-256, which has very, very, =
very, low odds of having the same value occur from two different blocks =
of data.

When ZFS dedupe originally came out you could have one of the following =
values:
	. off
	. on (=3D=3D sha256)
	. flecther4 with verify/compare
	. sha256 (without verify/compare)
	. sha256 with verify

There was a long-ish thread on zfs-discuss  fairly recently on whether =
SHA-256 was "good enough" where you could trust it, or whether one =
should do a verify step in addition to SHA-256:

	=
http://mail.opensolaris.org/pipermail/zfs-discuss/2011-January/046875.html=


While some people argued that it was prudent to use "verify" (especially =
with your data/job on the line), a good portion of folks though said =
that it's not worth it (i.e., if you're not worried about being hit by =
lightning (2^-17 to 2^-18), you shouldn't be worried about a hash =
collision (2^-128)).




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?525D503A-240C-49F2-9AAD-EC8E3C1CDB9A>