Re: Slower performance after enabling async io on Oracle Linux
From: Matthias Hoys <matthias.hoys_at_gmail.com>
Date: Tue, 15 May 2012 06:01:51 -0700 (PDT)
Message-ID: <18420171.890.1337086911679.JavaMail.geo-discussion-forums_at_vbli11>
On Monday, May 14, 2012 6:17:31 PM UTC+2, joel garry wrote:
> Well, it could be simply squeezing the bottleneck balloon, where
> something goes faster and consequently something depending on it slows
> down (for example, more cpu required to service more I/O). Some
> hugepages misconfiguration could have strange consequences simply from
> wasted memory, too.
>
> Also see, for example, http://blog.tanelpoder.com/2010/11/23/asynch-descriptor-resize-wait-event-in-oracle/
> fixed by your patch, but the linux/oracle async interaction is so
> delicate, you may be seeing something unfixed by the patch. Any
> errors like
> http://www.freelists.org/post/oracle-l/WARNINGio-submit-failed-due-to-kernel-limitations-MAXAIO-for-process128-pending-aio127,2
> ?
>
> jg
> --
> _at_home.com is bogus.
> http://www.utsandiego.com/news/2012/may/10/hackers-taking-aim-small-medium-businesses/
Date: Tue, 15 May 2012 06:01:51 -0700 (PDT)
Message-ID: <18420171.890.1337086911679.JavaMail.geo-discussion-forums_at_vbli11>
On Monday, May 14, 2012 6:17:31 PM UTC+2, joel garry wrote:
> Well, it could be simply squeezing the bottleneck balloon, where
> something goes faster and consequently something depending on it slows
> down (for example, more cpu required to service more I/O). Some
> hugepages misconfiguration could have strange consequences simply from
> wasted memory, too.
>
> Also see, for example, http://blog.tanelpoder.com/2010/11/23/asynch-descriptor-resize-wait-event-in-oracle/
> fixed by your patch, but the linux/oracle async interaction is so
> delicate, you may be seeing something unfixed by the patch. Any
> errors like
> http://www.freelists.org/post/oracle-l/WARNINGio-submit-failed-due-to-kernel-limitations-MAXAIO-for-process128-pending-aio127,2
> ?
>
> jg
> --
> _at_home.com is bogus.
> http://www.utsandiego.com/news/2012/may/10/hackers-taking-aim-small-medium-businesses/
Using vmstat/top, I don't see anything which could indicate a CPU bottleneck. I didn't configure hugepages, since this is a small Linux box with only 2GB SGA. When async io is enabled, I only see very few of the "asynch descriptor resize" waits. When filesystemio_options is set to NONE and disk_asynch_io is set to TRUE, the following warnings occur in archive logging trace files: "Log read is SYNCHRONOUS though disk_asynch_io is enabled!". There are no "maxaio kernel limitation" errors...
Matthias Received on Tue May 15 2012 - 08:01:51 CDT