Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 11 Jul 2012 08:32:34 -0700 (PDT)
From:      Jason Usher <jusher71@yahoo.com>
To:        Bob Friesenhahn <bfriesen@simple.dallas.tx.us>
Cc:        freebsd-fs@freebsd.org
Subject:   Re: vdev/pool math with combined raidzX vdevs...
Message-ID:  <1342020754.79202.YahooMailClassic@web122502.mail.ne1.yahoo.com>
In-Reply-To: <alpine.GSO.2.01.1207110829540.27589@freddy.simplesystems.org>

next in thread | previous in thread | raw e-mail | index | archive | help
=0AHi Bob,=0A=0A--- On Wed, 7/11/12, Bob Friesenhahn <bfriesen@simple.dalla=
s.tx.us> wrote:=0A=0A> The probabilty is indeed additive just as you say.=
=A0 My=0A> point is that the fundamental integrity is offered at the=0A> vd=
ev level.=A0 If a vdev fails, then the whole pool is=0A> gone.=A0 The MTTDL=
 calculations for various vdev=0A> topologies vary by orders of magnitude, =
which tends to make=0A> the additive nature of more vdevs insignificant.=0A=
=0A=0AThanks again for responding.=0A=0AI'm not going to beat this to death=
, but just to summarize, if F is 2, then the corresponding data loss probab=
ilities for RAID-Z1, -Z2, -Z3 are: 14.9%, 1.3%, and 0.086%.=0A=0ABut if com=
bining multiple vdevs into a zpool (as opposed to maintaining a different z=
pool for each raidz3 vdev) is additive, then raidz3 becomes .258%.=0A=0ASin=
ce (I think) a lot of raidz3 adoption is due to folks desiring "some overki=
ll" as they attempt to overcome the "disks got really big but didn't get an=
y faster (for rebuilds)"[1] ... but they are losing some of that by combini=
ng vdevs in a single pool.=0A=0ANot losing so much that they're back down t=
o the failure rate of a single raidz*2* vdev, but they're not at the overki=
ll level they thought they were at either.=0A=0AI think that's important, o=
r at least worth noting...=0A=0A=0A[1] http://storagegaga.com/4tb-disks-the=
-end-of-raid/



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1342020754.79202.YahooMailClassic>