Re: asynch descriptor resize wait event
Date: Tue, 01 Jun 2010 08:41:24 -0600
It turns out this was not the issue...severe performance degradation
due to optimizer_index_cost_adj and optimizer_index_caching were not set
Once that was fixed the waits disappeared. The incorrect settings were due to recommendations by a third party vendor.
SELECT IOUG Contributing Editor
Oracle Database 11g – Underground Advice for Database Administrators https://www.packtpub.com/oracle-11g-database-implementations-guide/book OCP 8i, 9i, 10g DBA
Southern Utah University
>>> Yong Huang <yong321_at_yahoo.com> 5/21/2010 9:22 AM >>>
If you decide to open an SR, please remember to post what the analyst says to the list, or at least email me. I'm curious. Thank you.
- On Thu, 5/20/10, April Sims <sims_at_suu.edu> wrote:
From: April Sims <sims_at_suu.edu>
Subject: Re: asynch descriptor resize wait event To: yong321_at_yahoo.com
Date: Thursday, May 20, 2010, 3:05 PM
>>> Yong Huang <yong321_at_yahoo.com> 5/20/2010 2:42 PM >>>
The only meaningful note on MOS I can find is 1081977.1. "This event is set when the number of asynch descriptor reserved inside the OS kernel has to be readjusted." On Linux, the number of system-wide AIO's is capped by /proc/sys/fs/aio-max-nr and its current value is aio-nr. I'm guessing after the said event occurs, /proc/sys/fs/aio-max-nr is raised. If so, and if the event happens often, it may be a good idea to manually raise aio-max-nr (echo some_large_number > /proc/sys/fs/aio-max-nr).
This undocumented wait event is recorded when there is an underlying IO descriptors (kiocb in linux?) needs to be adjusted. During the time all the pending IOs are suspended momentarily and resumed after the structure is expanded. I am not sure why this happens in an otherwise idle database. Perhaps a bug?
On Wed, May 19, 2010 at 3:48 PM, April Sims <sims_at_suu.edu> wrote:
> Anyone know what is causing "asynch descriptor resize" wait events
> 11gR2.(126.96.36.199)..on a test database which is mostly idle?
> Linux RH 4 - 64bit
> I can only find a limited amount of information on MOS and hardly
> in the Internet.
> Hanging in SQLPLUS when querying dba_autotask_client which led me to
> for wait events.
> April Sims