Comments on: Group by Hash aggregation

By: goldenorbit

goldenorbit — Thu, 02 Dec 2010 02:14:41 +0000

http://jonathanlewis.wordpress.com/2008/12/21/group-by/#comment-38013

I posted a question to Jonathan regarding the performance of “hash group by” operation in 10gR2 and 11g.

Jonathan started a poll to find out how many DBA actually disables this feature in 10g and 11g.

By: orainternals

orainternals — Thu, 07 Oct 2010 15:11:18 +0000

Hello Walter
Thanks for reading my blog. For some reason, your comments went in to spam folder, sorry about that.
I read that bug 8619631. Interesting issue. How much is the performance difference between the SORT GROUP BY and HASH GROUP BY in your case? Is this related to HASH GROUP BY using hashing method matching with partition boundaries? or something else? Would you please share some details?

Cheers
Riyaj

By: Walter

Walter — Thu, 07 Oct 2010 00:54:59 +0000

Riyaj,

Your post is interesting as we are suffering from the exact opposite, we want optimizer to perform HASH GROUP BY whereas it is performing SORT GROUP BY on our 11.1.0.7 database. Metalink has filed a bug for this as well

Note: 8619631.8

Walter.

By: orainternals

orainternals — Wed, 06 Oct 2010 21:41:23 +0000

Bhavik
I updated the blog with pstack/pmap commands. Hope that helps!
Cheers
Riyaj

By: Walter

Walter — Tue, 05 Oct 2010 21:08:32 +0000

Funny you mentioned this post as we are having the opposite issue on our 11.1.0.7 environment. Its mentioned in bug 8619361 and sort group by is taking an awful amount of time.

Walter.

By: orainternals

orainternals — Mon, 04 Oct 2010 18:35:03 +0000

Thank you for reading my blog.

Yasser
I think, I have a presentation about unix tools. I may not have uploaded yet, I will check up on it..

Bhavik
Thanks for sharing your experience. I guess, introducing a mandatory SORT option might avoid HASH GROUP BY feature in 10g and 11g. But, In future, Oracle might introduce grouping + sorting using HASH methods though.

Thanks
Riyaj

By: Bhavik Desai

Bhavik Desai — Mon, 04 Oct 2010 14:45:12 +0000

Hi Riyaj,

I encountered similar issue in one of my 10g data-ware house.
As you mentioned, HASH GROUP BY not bad and there are certainly some sets of sort mechanisms where HASH GROUP BY is better than SORT GROUP BY.
However, i resolved my issue by putting a dummy ORDER BY clause with existing GROUP by.
Note that, Order BY clause will mostly use SORT GROUP BY.

By: Yasser

Yasser — Sat, 02 Oct 2010 04:56:01 +0000

Excellent Scientific Approach !!

Could you please share pstack, pmap, and truss commands in detail, will be helpful for reference.

-Yasser