Re: CIO & flash recovery area

On 04/18/2006 07:51:44 PM, Allen, Brandon wrote:
> No replies yet to my initial inquiry, so I guess nobody is running their flash recovery areas on a cio filesystem. Oracle support referred me to the following link: http://www.rampant-books.com/t_oracle_direct_i_o.htm - which says "Oracle recommends that all database files use Direct I/O XE "Direct I/O" , which is a disk access method that bypasses the additional overhead on the OS buffer. One important exception to this rule is the archived redo log filesystem which should use OS buffer caching."
>
> Can anyone think of a reason why archived redo logs should be buffered instead of direct/concurrent I/O? It doesn't make sense to me.
>
> Thanks,
> Brandon

Brandon, why would that be important at all? Good news is that buffered I/O with prefetch usually exhibits much better performance on read then direct I/O. Bad news is that the only time when you are actually reading archived redo logs is when you are doing database recovery. That also covers a Data Guard configuration, in which one instance is always doing recovery. Oh, I forgot LogMiner. It, too, reads archived logs. If you don't have Oracle*Streams (LogMiner on steroids), DataGuard or something similar, the only time spent reading archived logs is a nice time spent doing database recovery. When you have to recover database while users are waiting and CIO (Chief Information Officer, not concurrent I/O) is looking over your shoulder, every nanosecond counts, trust me on that. To make long story short, archived redo logs aren't very important from performance point of view. You still have to back them up or your CIO (Career Is Over).

-- 
Mladen Gogala
http://www.mgogala.com

--
http://www.freelists.org/webpage/oracle-l