Oracle Linux IO Tunning - filesystemio_options- disk_asynch_io - PAGECACHE - aio-max-size - O_DIRECT

From: Chris Marquez <marquezemail_at_gmail.com>
Date: 2005-12-19 21:38:42
Message-id: d494e9760512191238we9dbd4eud51d68878283b711@mail.gmail.com

ENV:
Oracle 9205 -on- RHEL3 -on- EXT3

All,

I have been reading previous post, my own notes and some info on the internet.
I'm primarily interested in Oracle and using "filesystemio_options=asynch", but
feel all/many of these OS parameters / options are related, but not required?

I would like any and all opinions on these (below), but again I'm 99% sure I'm going with "filesystemio_options=asynch" ASAP.

---Oracle

---

• filesystemio_options=asynch|directio|setall|none

  ---

  1.) I know when, since 9iR2 and 10g...but *why* is asynch-IO not the default with Oracle RDBMS anymore? Wasn't it a monumental moment when in Oracle7?, Oracle Corp. made available Oracle RDBMS asynch-IO?...what we all wanted and

needed? I was a little surprised and caught off guard to find I had to re-implement and learn this stuff again?

I have a poor memory, but keep good notes...it seems I did implement this on my 9iRAC-RHEL3-OCFS system, but promptly un-did

it after obvious problems...it was recommended we use asynch-IO with OCFS. :o| Again for this thread I'm on *EXT3* now.

---

• disk_asynch_io=TRUE

  ---

  2.) I guess this is obvious...if I use filesystemio_options=asynch, would leave the default "disk_asynch_io=TRUE" setting.

---OS/Linux

---

• PAGECACHE

  ---

  3.) Never used it. These seems to controls my/the OS IO cache that Oracle filesystemio_options=asynch would use? I have the OS default and scared to make changes with know the true impact.

[oracle_at_db08 bdump]$ more /proc/sys/vm/pagecache 1 15 100

max=100(%)...I think I like knowing that OS will not "move process pages to swap"?

http://www.redhat.com/magazine/001nov04/features/vm/ pagecache
The pagecache file adjusts the amount of RAM which can be used by the page cache.

---

• aio-max-size

  ---

  4.) Never used it. This seems like good info...I have 2 OLTP db's and 1 OLAP db. Anyone have experience with these (when using Oracle asynch I/O)?

http://www.puschitz.com/TuningLinuxForOracle.shtml

Increasing I/O Throughput at the Linux OS Level

The /proc/sys/fs/aio-max-size parameter can be changed if asynchronous I/O is used for Oracle datafiles residing on filesystems (e.g. "ext2"). ...
For Online Transaction Processing (OLTP) workloads, the default size of 131072 would suffice.

---

• O_DIRECT

  ---

  5.) This seems obvious and only at the OS level...this is covered (in Oracle) by the filesystemio_options parameter.

I await your input...Thanks!

--
Chris Marquez
Oracle DBA

Received on Mon Dec 19 2005 - 21:38:42 CST