Date: Tue, 26 May 2009 22:38:47 +0300 From: Artis Caune <artis.caune@gmail.com> To: peterjeremy@optushome.com.au Cc: Randy Bush <randy@psg.com>, freebsd-fs <freebsd-fs@freebsd.org> Subject: Re: raidz2 a bit big Message-ID: <9e20d71e0905261238x2ff5a96cu8891f00c802a0acc@mail.gmail.com> In-Reply-To: <20090526185900.GA98171@server.vk2pj.dyndns.org> References: <m28wkon90c.wl%randy@psg.com> <9e20d71e0905230537ibcaf852g1dc32b6ffc3a681d@mail.gmail.com> <20090526185900.GA98171@server.vk2pj.dyndns.org>
next in thread | previous in thread | raw e-mail | index | archive | help
2009/5/26 <peterjeremy@optushome.com.au>:
> On 2009-May-23 15:37:14 +0300, Artis Caune <artis.caune@gmail.com> wrote:
>>2009/5/23 Randy Bush <randy@psg.com>:
>>> a dozen 2tb drives in a raidz2
>>Reads on such configurations are very slow.
>
> Not really. =C2=A0Assuming each disk is capable of X IOPS and assuming
> non-degraded mode, you can still issue (N-2)*X random reads/sec
> because the parity stripes are not needed/used. =C2=A0(Compared to N*X
> random reads/sec for a mirrored configuration). =C2=A0Degraded reads _are=
_
> very slow because you need to read most of the spindles (I'm not sure
> of the exact recovery mechanism for RAIDZ2 but it's probably close to
> X random reads/sec).
Not really,
RAID{5,6} !=3D RAIDZ{1,2}
http://blogs.sun.com/roch/entry/when_to_and_not_to
http://utcc.utoronto.ca/~cks/space/blog/solaris/ZFSRaidzReadPerformance
We switched 20K mailboxes from raidz2 to mirror because of slow reads.
--=20
Artis Caune
Everything should be made as simple as possible, but not simpler.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?9e20d71e0905261238x2ff5a96cu8891f00c802a0acc>