Re: Average physical read times for AIX & SAN

I confess I missed this as well. We had an identical problem that went beyond astronomical times for log writes, it included write times similar to those displayed below on data files. The issue threatened to cause cancellation of an entire major project.

Our solution was pretty simple. Check the segment size on the array. In our case, every 'expert' (consultant) insisted 8k was optimal. Nonsense. When we finally bumped it up to 64k and moved the logical volumes containing the redo logs off the disks on which the data files were striped, throughput went to acceptable levels.

Average ms for reads on our datafiles is 2.

On Thu, Oct 9, 2008 at 2:29 PM, Claudia Zeiler <czeiler_at_ecwise.com> wrote:

> Daniel,
> I am sorry to be so slow to respond to this, it was caught by my junk
> filter and only emerged today.
>
> Reading this, at first I thought that you were internal to my company. We
> have the same configuration and here are a few gem lines from
> sid_lgwr_123.trc
>
> Warning: log write time 620ms, size 2501KB
> *** 2008-10-08 18:13:50.581
> Warning: log write time 2000ms, size 9435KB
> *** 2008-10-08 18:13:51.779
> Warning: log write time 870ms, size 4072KB
> *** 2008-10-08 18:13:53.07
>
> I have seen these write times also when less than 1KB was being written.
> My winner warning so far was for 23 seconds!
> If you have learned anything about this issue, I would LOVE to hear it.
> Thanks a million,
> Claudia Zeiler
>
> -----Original Message-----
> From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org]
> On Behalf Of Daniel Fink
> Sent: Wednesday, September 24, 2008 9:18 AM
> To: oracle-l
> Subject: Average physical read times
>
> In a recent optimization effort, I found what I think are high physical
> read times. For example, a single block read was averaging 11
> milliseconds with a max of over 1 second (tkprof output). For this
> session, single block reads consumed over 80% of the response time, so
> this is of concern.
>
> We are running Oracle 10gr1 on AIX with an IBM storage array (shared
> amongst many servers). I suspect there is some optimization that can be
> done at the storage array level. If 11 milliseconds is well above
> average, this will help in engaging that team in the process.
>
> For those of you running statspack. AWR or other monitoring tools, what
> are the average single block read times you are encountering on a good
> system? If you are running on an IBM storage array, have you found any
> specific issues/fixes?
>
> Regards,
> Daniel Fink
>
> --
> Daniel Fink
>
> OptimalDBA.com - Oracle Performance, Diagnosis, Data Recovery and Training
>
> OptimalDBA http://www.optimaldba.com
> Oracle Blog http://optimaldba.blogspot.com
>
> Lost Data? http://www.ora600.be/
>
> --
> http://www.freelists.org/webpage/oracle-l
>
>
> --
> http://www.freelists.org/webpage/oracle-l
>
>
>

--
http://www.freelists.org/webpage/oracle-l