Comments on: METHOD_OPT

By: Statistics poll « Timur Akhmadeev's blog

Statistics poll « Timur Akhmadeev's blog — Mon, 16 Apr 2012 19:55:36 +0000

[...] to everyone for spreading the link and participating in the poll on statistics. As it was pointed out in the comments, the poll isn’t very specific, I know. [...]

By: albertofrosi

albertofrosi — Wed, 11 Apr 2012 14:58:14 +0000

Hi Timur,
I would prefer to use my brain, least as long as it works…
I’m very curious for this interesting poll.
I will wait the result…:-)
thanks
Ciao

Alberto

By: savvinov

savvinov — Wed, 11 Apr 2012 10:31:35 +0000

Hi Timur,

no I haven’t.

Best regards,
Nikolay

By: Timur Akhmadeev

Timur Akhmadeev — Wed, 11 Apr 2012 10:10:46 +0000

Hi Kellyn

me too but at the time of creating the poll I was thinking about the METHOD_OPT mostly, and it’s not much difference from user perspective in this area. You either get histograms when Oracle thinks it is needed or you don’t trust Oracle and use your brains instead.

By: Timur Akhmadeev

Timur Akhmadeev — Wed, 11 Apr 2012 10:07:25 +0000

Thank you for your feedback.

but I suspect it would have been many more if I were using method_opt=>’… size 1′.

Have you tried it though?

By: Timur Akhmadeev

Timur Akhmadeev — Wed, 11 Apr 2012 09:59:29 +0000

Thank you for your feedback.

By: Timur Akhmadeev

Timur Akhmadeev — Wed, 11 Apr 2012 09:58:36 +0000

Thanks for comment. As far as I know automatic job does not gather fixed objects statistics – but I might be wrong and have to check Maria’s paper

By: Kellyn Pot'Vin (@DBAKevlar)

Kellyn Pot'Vin (@DBAKevlar) — Tue, 10 Apr 2012 15:45:40 +0000

I’d be curious how the results would change if the poll differentiated between 10g and 11g for this question. Thanks!

By: savvinov

savvinov — Tue, 10 Apr 2012 12:18:09 +0000

I haven’t worked with 11g much, but 10g default options seem adequate in most cases. Although I did have one major incident because of method_opt=>’…size auto’ (a histogram was generated where not appropriate) but I suspect it would have been many more if I were using method_opt=>’… size 1′.

By: CJ

CJ — Tue, 10 Apr 2012 11:38:05 +0000

It is probably worth clarifying whether this includes stats gathering on data dictionary and fixed tables or not.

I would leave the default job gather the above stats, while creating a custom framework for stats on application objects.

By: lscheng

lscheng — Tue, 10 Apr 2012 09:55:11 +0000

The default is ok with my customer environment in 10gR2 and 11gR1, 11gR2. In 10gR2 we had to increase degault percentage to 30 and had to do some custom statistics gathering for a few tables but most are ok

By: Timur Akhmadeev

Timur Akhmadeev — Tue, 10 Apr 2012 07:43:09 +0000

Hi Tim

thank you for your feedback.

I would expect far fewer people to be using the defaults in 10g compared to 11g.

Very good point. I agree 11g defaults are different especially in account to SAMPLE_SIZE.

Another thing I didn’t thought about thoroughly is how “custom” custom stats management could be. It could be just a matter of time frame (so defaults are still there, but job runs more frequently, for example) – and for me this case is equal to the first poll option. What I mean by “custom stats management”: not gathering statistics (plus maybe copying (sub)partition stats); custom statistics manipulation; DBMS_STATS calls integrated in the application/maintenance scripts – all the ways that do not rely on the job.

By: oraclebase

oraclebase — Tue, 10 Apr 2012 07:28:37 +0000

Given a sufficient amount of time and a variety of systems being worked on, I would expect even the laziest DBA to have all three methods. On low priority systems the defaults are fine in many cases. Some systems need a little customizing and some edge cases are so extreme that defaults really don’t work well.

Perhaps the poll should have mentioned a specific version also. I would expect far fewer people to be using the defaults in 10g compared to 11g.

Cheers

Tim…

By: Stats Collection « Oracle Scratchpad

Stats Collection « Oracle Scratchpad — Tue, 10 Apr 2012 07:13:39 +0000

[...] is just a temporary note to point you to a poll set up by Timur Akhmadeev to get an idea of how people are handling stats collection in newer versions of [...]

By: Optimizer Statistics Whitepaper « Anand's Blog

Optimizer Statistics Whitepaper « Anand's Blog — Mon, 09 Apr 2012 20:00:01 +0000

[...] http://timurakhmadeev.wordpress.com/2012/04/09/method_opt/ Share this:TwitterLike this:LikeBe the first to like this post. [...]