RE: How would you layout the files?
Date: Mon, 8 Dec 2008 16:44:19 -0600 (CST)
You do this Greg, (and I don't argue it's viability) as long as you manage client expectations and make them aware that they could lose EVERYTHING in the event of a disk failure (that's including in-flight transactions in redo, and logs which haven't switched), as opposed to perhaps merely needing to recover a lost datafile on a single disk which goes bad (of course, if that single disk contains those files - oh, well).
The benefit of your SAME solution is you have hardware redundancy. The only downside is I've seen raided stuff go awry and it's very messy if and when it does happen.
Back in the day, Oracle's solution was a sort of software mirroring one - you'd duplex important files (such as online redo, controlfiles, software), and then recover the rest if they were lost. This is still recommended to keep human error at bay!
Lest we forget, I bet we can find several posters still running Oracle 7 on 10 year old hardware on this list, with a configuration very similar to what I've just described. I recall finally retiring one for a client just last year....
From: Greg Rahn [mailto:greg_at_structureddata.org] Sent: Monday, December 08, 2008 4:36 PM
Cc: moabrivers_at_gmail.com; oracle-l_at_freelists.org Subject: Re: How would you layout the files?
SAME - stripe and mirror everything.
Redo and Data should be fine together. In fact, with only 4 spindles available for the db, I wouldn't have it any other way. RAID-10 the 4 drives and mount it up as a single volume. Pray for the best.
The problem that people create by separating redo, is they limit the number of spindles and thus reduce the IOPS that can be supported. While often one spindle is enough for space, it is rarely enough in terms of performance (IOPS).
On Mon, Dec 8, 2008 at 1:57 PM, Michael Fontana <mfontana_at_enkitec.com>
> Being that this appears to be a 2002 solution, it's not so bad. I would
> look at your OFA guidelines for disk placement in an 8i administrators
> guide. You're going to want to keep the redo away from the datafiles,
> make sure you have multiple copies of same. Do all that and then pray
> don't have a significant hardware failure, all the while, having good
> recovery procedures in place with measurably tested results; inform the
> client as to expectations along those lines, as well as performance, and
> you'll be fine.
-- Regards, Greg Rahn http://structureddata.org -- http://www.freelists.org/webpage/oracle-lReceived on Mon Dec 08 2008 - 16:44:19 CST