Re: Configuring RAID for best Oracle performance

We have been testing various disk configurations including some RAIDs. Our tests include an OLTP type of application which is write intensive.

In our tests so far, we have seen at least a 15% performance degradation when the redo logs were on a RAID level 3 device. Our testing will include other RAID levels but at the moment only RAID level 3 has been tested. If performance is what you are after (and who isn't), the redo logs should not be placed on anything other than a RAID 1 (shadowing). (It will be interesting to see if RAID 5 can handle the redo logs any better but I have my doubts.) We have not tested RAID 1 in comparison with Oracle mirroring but common sense indicates that the hardware should perform better.

Also, with one of the drives in the RAID placed offline, we saw a 7% performance degradation. Of course, these numbers will vary with different RAID vendors but they should be in the same ballpark within a RAID level.

So far, if redo logs are kept off of the RAID 3 device, the database performance is as good as having the database split across the same number of disk drives that comprise the RAID - 1. (I am subtracting one because in RAID level 3, there is a parity drive which does not have real data written to it but only XOR information used to produce the correct data in the event of a read failure.) Hence, with no or a very small performance hit, you can have some redundancy.

More information as it is created:-)

Walter

• The opinions written here are mine and mine alone.