Re: RAID Storage for Oracle

Striping data *increases* performance, if properly configured. RAID5 sets' penalty is due to the parity information that must be distributed across the RAIDset members, along with the data. Many controllers have enough cache to be able to buffer this activity, however, and lessen the immediate effects of this overhead, assuming no cache flooding.

Do not assume that just because RAID0+1 ultimately performs best, RAID5 won't do just as well in a particular configuration. You're probably not trying to achieve 100,000 TPmC-C -- you're trying to run a business within a budget. It depends on what the system is doing, what it's designed for, how well the application is written and how well the database transactions are coded, who's going to be using it and for what purposes, what else is going to be running on the system and when, and blah blah blah.

By the way, I often software-mirror controller-based stripesets and it can be extremely effective. For example, I put together a DEC AlphaServer 1200 running Digital UNIX, with 2 Mylex controllers, each containing a 6-way stripe. Talk about fast! It was a HR transaction system and people couldn't believe it was running on this tiny little box shoved under a 6-foor folding table.

I've also put togther terabyte-sized systems that used ONLY RAID5 except for the redo logs which were 3-way mirrors, the reason being that if the database died you were in serious doodoo, a CIO or datacenter manager willing to accept the additional cost of lots and lots of RAID5 sets and their attendant controllers (for bandwidth) in trade for reliability.

Hth,
WJD

• Sent from RemarQ http://www.remarq.com The Internet's Discussion Network * The fastest and easiest way to search and participate in Usenet - Free!