Re: AIX Concurrent I/O for Oracle archive logs

  • From: "dbvision@xxxxxxxxxxxx" <dbvision@xxxxxxxxxxxx>
  • To: oracle-l@xxxxxxxxxxxxx, 'Rich Jesse' <rjoralist@xxxxxxxxxxxxxxxxxxxxx>
  • Date: Thu, 27 May 2010 07:26:22 +0800

On Wed May 26 21:46 , "Rich Jesse"  sent:

>
>Does the "Tuning IBM AIX 5L for an Oracle Database" PDF (Google it -- my
>copy has the intuitive filename of "9a46.pdf") qualify as one of these
>papers?

If it's got Dennis Massenari's name in it, it's likely worth your time reading.
The guy and his team are doing excellent work in documenting how to configure 
AIX
for Oracle.  Look for the latest stuff. It's still waaaay too complex and
confusing IMHO, but it's hugely better and more reliable than what we used to 
have!

The problem I find is that some of the older, non-qualified AIX tuning stuff
remains available and can get folks confused.

>Same with the MOS article and the IBM paper I mentioned.  Is CIO mounting of
>the archives going to slow something down?  Or perhaps potentially relieve
>swap pressure on a system with low RAM and an auto-adjusting file cache,
>resulting in an overall performance increase?


Well, part of the problem is that turning CIO on by itself is not enough.  You
also have to tune the memory management with "vmo" to make AIX use less memory
for file cache and more for programs/SGA.  
On the other hand, if you allocate largepages for SGA and Oracle and let Oracle
10gr2 turn on CIO where needed, you are better off letting vmm use more file
cache for what remains of physical memory after Oracle grabbed all largepages:
that way you can optimize file cache for all files not used with CIO and let
Oracle use CIO and SGA direct IO only for the files where it really matters. 
That is my preferred solution, anyway.  I wrote a lot on this in this blog entry
(apologies for the title!), go through the lot and comments as well - good stuff
there:
http://dbasrus.blogspot.com/2009/06/size-sometimes-does-matter.html


>
>Without a definitive answer on the ramifications of the action/inaction, it
>sounds like empirical evidence and a good DBA's best judgement are the key
>tools.  Then again, they usually are...  :)

You got it in one!  Couldn't agree more.

Cheers
Nuno
--
//www.freelists.org/webpage/oracle-l


Other related posts: