From: Chris Browne <cbbrowne@acm.org>
Subject: Re: Advice on RAID card
Date: Sun, 25 Sep 2005 20:10:00 -0400
Organization: cbbrowne Computing Inc
Lines: 26
Message-ID: <60ek7cn0yv.fsf@dba2.int.libertyrms.com>
References: <IGEGINLEEDLANNHAAKCMOECJCCAA.gaulakh@ecmarket.com>
	<op.sxll3du6th1vuj@localhost> <43367937.40809@canaan.co.il>
	<ABE6B997-2C1E-43B1-9F18-28F54B1620FE@fastcrypt.com>
	<4336C57B.6090800@commandprompt.com> <op.sxn1sdofth1vuj@localhost>
	<4336CEAA.7070208@commandprompt.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
User-Agent: Gnus/5.1007 (Gnus v5.10.7) XEmacs/21.4.17 (Jumbo Shrimp, linux)
Cancel-Lock: sha1:1KLdeeRKSnLlTLkWp2Oj1YAuCyg=
To: pgsql-performance@postgresql.org

jd@commandprompt.com ("Joshua D. Drake") writes:
> There is a huge advantage to software raid on all kinds of
> levels. If you have the CPU then I suggest it. However you will
> never get the performance out of software raid on the high level
> (think 1 gig of cache) that you would on a software raid setup.

This appears to be a case where the "ludicrous MHz increases" on
desktop CPUs has actually provided a material benefit.

The sorts of embedded controllers typically used on RAID controllers
are StrongARMs and i960s, and, well, 250MHz is actually fast for
these.

When AMD and Intel fight over adding gigahertz and megabytes of cache
to their chips, this means that the RAID work can get pushed over to
one's "main CPU" without chewing up terribly much of its bandwidth.

That says to me that in the absence of battery backed cache, it's not
worth having a "bottom-end" RAID controller.  Particularly if the
death of the controller would be likely to kill your data.

Battery-backed cache changes the value proposition, of course...
-- 
select 'cbbrowne' || '@' || 'acm.org';
http://cbbrowne.com/info/linuxdistributions.html
All generalizations are false, including this one.