Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Is Raid 5 really that bad for Oracle?

Re: Is Raid 5 really that bad for Oracle?

From: Howard J. Rogers <hjr_at_dizwell.com>
Date: Sun, 1 Aug 2004 09:07:25 +1000
Message-ID: <410c2603$0$2542$afc38c87@news.optusnet.com.au>

"joe bayer" <joebayerii(no-spam)@hotmail.com> wrote in message news:qDPOc.1753$%J6.1677_at_trndny07...
> I am quoting from Jonathan Lewis's book, Practical Oracle 8i, page 206
>
> Raid 5 has an undeservedly bad reputation as far as Oracle database
systems
> are concerned. ....
> However, for most small systems, it is almost necessary and perfectly
> acceptable; and for many large systems it is totally adequate.
> ....
> The first point is that many Oracle systems do far less writing than they
do
> reading, and the writing is usually in the backgroud anyway, so although
the
> write penalty is a notional 100%, this in=s not necessarily all that
> significant and overhead to the total I/O operation and many not impact
the
> user directly anyway.
>
> Second, although a single , small random write is likely to sustain a
heavy
> penalty, RAID 5 suppliers are aware of the issue and have taken steps to
> reduce the problem. Typically a write that fills a whole stripe does not
> need to re-creates the parity as it writes each disk. It simply discards
> and re-creates a new parity block.
>
> Third, it is beoming common practice to stick a reasonably large
> battery-backed cache on the Raid 5 device.

Sure, RAID-5's write penalty is not a problem for DBWn, because that process does background writes anyway (though if it gets too sluggish, you could be in for free buffer waits). But LGWR writing to RAID-5 is definitely not a good idea: a commit's not a commit until LGWR can confirm it's written the relevant redo to disk, and if there's a write penalty to negotiate there, your users will wait on it.

But the write penalty is only one of RAID-5's problems, and perhaps not the major one at that (especically given the prevalance of battery-backed caches as you mention). The real issue it seems to me is performance when one of the RAID members dies (I/O against the entire stripe set to perform one physical read is not such good news). And a subsidiary matter is that the "I" of RAID used to mean 'inexpensive', and as you can read at www.baarf.com RAID-5 is not so inexpensive as one might think.

Regards
HJR Received on Sat Jul 31 2004 - 18:07:25 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US