Re: Concurrent io AIX 5.3

  • From: "Alex Gorbachev" <ag@xxxxxxxxxxxx>
  • To: bshan@xxxxxx, oracle-l@xxxxxxxxxxxxx
  • Date: Thu, 26 Jul 2007 19:04:29 -0400

I second opinion that controlfiles can also be on CIO. Can you share
your conflicting references?

For redo logs - IO size is usually higher than one block. Read IO can
be up to 1MB don't know if AIX platform limits it. Write size depends
on how much of redo buffer LGWR needs to flush. LGWR does write in
chunks of 512 bytes or whatever your redo block size is but it issues
single write IO up to max it can.

On 7/26/07, Brian Shanblatt <bshan@xxxxxx> wrote:
Hello all!

I am configuring a new AIX 5.3 machine, for an oracle database

The file systems are JFS2, with the blocksize of 4K which matches our db
block size.

I would like to implement Concurrent io.  I am looking for some
advice/affirmation, beyond what I found on Metalink and IBM whitepapers.

What I would like to do is:
1) All datafiles (tables, indexes, UNDO tablespace, etc) would be on JFS2
mounted with cio, with a block size of 4K

2) Redo logs would be on JFS2, mounted with cio, with block size of
(agblksize) of 512 BYTES.  (is this because oracle writes redos in 512
byte pieces?)

3) Control files would be on on JFS2 file systems, mounted as default (not
using direct io or concurrent io).

Does this makes sense? I am assuming that control files should not be on
filesystems with cio, though I found some conflicted data about control

I would appreciate any advice/affirmations/flames!

Thank you!
-brian s


Alex Gorbachev, Oracle DBA Brewer, The Pythian Group
BAAG party -

Other related posts: