> -----Original Message----- > From: Dave Ewart [mailto:[EMAIL PROTECTED] > Sent: 30 August 2007 16:55 > To: [email protected] > Subject: Re: Software vs Hardware RAID 10? > > On Thursday, 30.08.2007 at 10:37 +0200, Goswin von Brederlow wrote: > > > > [RAID 6] > > > > If you write a single block then you have to read the 2 > parity blocks > > for the stripe, update them and write 3 blocks. So you have 5 times > > the traffic on the bus. For anything doing single block > writes this is > > quite the killer. I would imagine a database like oracle to > just die > > with software raid6. > > *nods* I believe most database 'vendors' recommend RAID-10, > and specifically to avoid any of the parity-based RAIDs, such > as RAID-5 and RAID-6. Presumably for exactly the reasons > you've outlined. For example, see > http://edoceo.com/liber/db-postgresql-performance.php
No, the main reasons for recommending against RAID-5/6 are not connected to performance during normal operation. Choosing the correct storage system is never primarily about speed, the first considerations are always: reliability, fault-tolerance, disaster-recovery, and fool-proof setup. The Battle Against Any Raid Five initiative provides much more information on why RAID-5 and RAID-6 should be avoided for production servers: <http://www.miracleas.com/BAARF/BAARF2.html?30,42> The above site should be required reading for anyone with critical data. Paul
