Re: AIX Concurrent I/O for Oracle archive logs
- From: Nuno Souto <dbvision@xxxxxxxxxxxx>
- Date: Wed, 26 May 2010 21:17:48 +1000
Allen, Brandon wrote,on my timestamp of 26/05/2010 2:06 AM:
Can anyone think of a good reason why Oracle’s archived redo logs
shouldn’t be placed on a non-buffered (concurrent or direct IO) file
system? Oracle Support document 418714.1 specifically states NOT to use
DIO/CIO for file systems containing archive logs. However, I have one
AIX system where my archive logs have been on a file system mounted with
CIO for years and I’ve never had any problems (and yes, I’ve
successfully used the logs for actual recovery).
As in everything, we have to put some context in the answers.
Please turn on monospaced font to see below:
IF - you are running >= AIX 5.2L
IF - you are running >= Oracle 10g
IF - all your file systems are JFS2 (and I mean JFS2, not JFS!)
THEN - you do not need to mount with CIO:
simply set filesystemio_options = setall
and let Oracle turn on CIO where it needs/wants.
ELSE - convert everything to JFS2:
JFS does not support CIO w/Oracle
ELSE - isn't it time you upgraded?
Well then: if not, use JFS2 and mount with CIO, no prob.
ELSE - about time you upgraded everything...
But go ahead and use JFS, you won't be able to use JFS2.
That means no CIO for you!
Hope the above is more or less clear with the code-like indents.
I’m considering moving another of my systems to CIO because on the other
system, when archive log backups run, the AIX file system cache gets
flooded by the archive logs and it has recently started causing the OS
to start paging heavily, which in turn is bringing the system to a
screeching halt for a few minutes, so I think if I move the logs to a
CIO file system then they should no longer be read into the AIX buffer
cache, and that should prevent this problem, but I just want to see if
anyone has any good reasons why I should avoid using CIO with archive logs.
Reads like good solid logic thinking to me.
Have you/are you using AIO? That might help for all IO, not just CIO originated
from db or f/s.
One caution: AIX by default does not configure enough started AIO server
processes to satisfy Oracle's demands. Jack it up to at least twice the total
number of datafiles. That's how many you need started to have top IO.
No, they do NOT consume a tremendous amount of resources: they are shared code
with very little data area and are only active for the duration of the IO,
mostly in wait state for the AIX scheduler.
Above all: be very careful with the doco you find online. There are perhaps two
documents in the entire IBM tech lib that are relevant and pertinent for current
versions of software and OS.
Make sure the date of any doc you read on this is *actual*, *and* relevant for
your OS and Oracle levels.
If it doesn't mention that information, simply ignore the document: it'll only
cause grief. (Once burnt twice shy, here...)
--
Cheers
Nuno Souto
in wet Sydney, Australia
dbvision@xxxxxxxxxxxx
--
//www.freelists.org/webpage/oracle-l
Other related posts:
