Re: Redo logs a Quick I/O enabled file system

On Thu, 13 Sep 2007 12:38:43 -0700, fitzjarrell_at_cox.net wrote:

> No 'impact', really, but no benefit, either, as they are written
> serially. I found no reason to place redo logs on such a file system.
>

What does quick I/O (aka "direct I/O") have to do with the files being written serially? Oracle will bypass data buffers when writing to redo logs which it does anyway, regardless of the direct I/O being present or not. When using direct I/O, an additonal buffer flush will be avoided which makes writing to redo logs a little bit faster, because of avoiding a context switch. On Linux platforms you can use ltrace command to trace library functions like fflush to verify that statement. Redo logs are written sequentially, in the same fashion as tapes used to be written and the session that issued a commit is waiting until that write is performed (at least, it used to with "COMMIT WORK". Oracle 10g formalized "asynchronous commit" with things like "COMMIT WRITE BATCH NOWAIT" or "COMMIT ARE YOU FREAKING KIDDING ME"). Of course, buffered asynchronous "commit" sort of defeats the purpose and is used to formalize the kind of commit processing done within PL/SQL loops.

-- 
http://www.mladen-gogala.com