Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 01 Feb 1998 00:09:05 +0000
From:      Thordur Ivarsson <totii@est.is>
To:        Brian Tao <taob@nbc.netcom.ca>
Cc:        Karl Denninger <karl@mcs.net>, current@FreeBSD.ORG
Subject:   Re: RAID controllers - folks, check this thing out
Message-ID:  <34D3BD21.D2E76C22@est.is>
References:  <Pine.GSO.3.95.980131170825.27817a-100000@tor-adm1>

next in thread | previous in thread | raw e-mail | index | archive | help
Brian Tao wrote:
> 
> On Sat, 31 Jan 1998, Karl Denninger wrote:
> >
> > RAID 5, due to the way it stripes parity across the volumes, has a
> > "sweet spot" in performance at 5 spindles.
> 
>     What "way" is that?  On a given stripe, one drive provides the
> parity block, the choice of drive staggered across consecutive
> stripes.  There may be an issue with small, sequential writes on a
> RAID 5 set with a large number of drives, but I can't think of any
> reason why five drives should be magical.
> --
> Brian Tao (BT300, taob@netcom.ca)
> "Though this be madness, yet there is method in't"

You should read the RAIDframe papers at
http://auchentoshan.pdl.cs.cmu.edu/RAIDframe/
the algorithm makes the difference in small writes, the small write has
to be broken up into smaller units to spread to 'stripe width' of disks,
or append to already written data on disk therefore 'read - append -
write' sequence in stead of just 'write'.

At least is this what I got out of those papers and what I have been
told of RAID specialists.

Thordur Ivarsson



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?34D3BD21.D2E76C22>