RE: Testing Direct I/O

>>In order to measure the performance impact of switching to direct I/O, it would
be a good thing to measure your performance, in the first place. Completely illogical
thing to do would be to develop a metrics to test your performance, then run a test
to create a baseline against which you are testing, then switch to direct I/O, rerun
the test and compare the results against the baseline. Unfortunately, my ESP powers
are usually weak before summer so I cannot read your mind and create a test that
would reflect your business model. My telepathic abilities increase exponentially
if I am plied with copious quantities of Long Island ice tea, but as that is not the
case, I can not help you.<<<

Well that is part of what I'm asking the list...

Are there specific metrics to quantify the gain in switching to direct i/o? What should be inspected? Some specific Oracle wait analysis, OS analysis, etc...

I'm not asking for anyone to read my mind but rather for what should be observed during testing...

By observing and documenting the results of a test I can try to understand the results better...I do-not want to blindly implement direct i/o without understanding it in my environment first(that was the message I was trying to convey)...

That is what this whole sentence was about:

"However, I'd like to test the results of actually using direct i/o. In order to not re-invent the wheel, I was wondering if anyone had performed similar testing?"

-----Original Message-----
From: Mladen Gogala [mailto:gogala_at_sbcglobal.net] Sent: Monday, February 13, 2006 4:14 PM
To: Khemmanivanh, Somckit
Cc: oracle-l_at_freelists.org
Subject: Re: Testing Direct I/O

On 02/13/2006 06:11:47 PM, Khemmanivanh, Somckit wrote:

>
> I saved and executed a few benchmark SQL statements pre-direct i/o, is
> this enough? Can I simply re-exexcute these benchmark statements?

Yes you can. Make sure those statements are laced with several "TRUNCATE" and
"DROP" statements as they are the best for checking the benefits of direct I/O.

>
> What metrics should I be focusing on to "quantify" direct i/o benefit
in
> my environment?

Well, the answer that is always correct is: "that depends". My favorite metrics in
the winter months is the number of calories per serving. Alternatively, you can
try with the BCHR. Allegedly, that is a good overall number which tells you how
well your system would be performing had it not been for those pesky solar flares.

>
> What I don't want to do is switch to direct i/o and have performance
> decrease (i.e. temp TS space was de-buffered) or not understand
what/why
> performance decreased...

In order to measure the performance impact of switching to direct I/O, it would
be a good thing to measure your performance, in the first place. Completely illogical
thing to do would be to develop a metrics to test your performance, then run a test
to create a baseline against which you are testing, then switch to direct I/O, rerun
the test and compare the results against the baseline. Unfortunately, my ESP powers
are usually weak before summer so I cannot read your mind and create a test that
would reflect your business model. My telepathic abilities increase exponentially
if I am plied with copious quantities of Long Island ice tea, but as that is not the
case, I can not help you.

>

-- 
Mladen Gogala
http://www.mgogala.com



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