Comments on: Does the Query Optimizer Cost PX Distribution Methods? http://antognini.ch/2010/01/does-the-query-optimizer-cost-px-distribution-methods/ Fri, 10 Sep 2010 07:45:48 +0000 http://wordpress.org/?v=2.9.2 hourly 1 By: Parallel query distribution methods « Tony’s Oracle Tips http://antognini.ch/2010/01/does-the-query-optimizer-cost-px-distribution-methods/comment-page-1/#comment-8275 Parallel query distribution methods « Tony’s Oracle Tips Wed, 10 Mar 2010 16:47:21 +0000 http://antognini.ch/?p=45#comment-8275 [...] asked Christian about this and his full reply came in this blog. Christian was able to explain why the estimated costs in the execution plan were misleading and [...] [...] asked Christian about this and his full reply came in this blog. Christian was able to explain why the estimated costs in the execution plan were misleading and [...]

]]> By: Christian Antognini http://antognini.ch/2010/01/does-the-query-optimizer-cost-px-distribution-methods/comment-page-1/#comment-8101 Christian Antognini Wed, 03 Mar 2010 07:57:44 +0000 http://antognini.ch/?p=45#comment-8101 Hi Michael Thank for the interesting information. Unfortunately the base bug, 6714199, is not public. Anyway, this is an addition evidence that there are problems in this area. Cheers, Chris Hi Michael

Thank for the interesting information. Unfortunately the base bug, 6714199, is not public. Anyway, this is an addition evidence that there are problems in this area.

Cheers,
Chris

]]> By: Michael http://antognini.ch/2010/01/does-the-query-optimizer-cost-px-distribution-methods/comment-page-1/#comment-8089 Michael Tue, 02 Mar 2010 23:43:16 +0000 http://antognini.ch/?p=45#comment-8089 I've seen problems with combined partitions and PWJ as in Tony Hasler's test. In my case it was caused by bug 6476257. Another requirement was that default DOP of a table should match the number of subpartitions. I’ve seen problems with combined partitions and PWJ as in Tony Hasler’s test. In my case it was caused by bug 6476257.
Another requirement was that default DOP of a table should match the number of subpartitions.

]]> By: Blogroll Report 22/01/2009 – 29/01/2010 « Coskan’s Approach to Oracle http://antognini.ch/2010/01/does-the-query-optimizer-cost-px-distribution-methods/comment-page-1/#comment-7880 Blogroll Report 22/01/2009 – 29/01/2010 « Coskan’s Approach to Oracle Sun, 21 Feb 2010 03:26:30 +0000 http://antognini.ch/?p=45#comment-7880 [...] 19-How does CBO behave with different PX Distribution Methods ? Christian Antognini-Does the Query Optimizer Cost PX Distribution Methods? [...] [...] 19-How does CBO behave with different PX Distribution Methods ? Christian Antognini-Does the Query Optimizer Cost PX Distribution Methods? [...]

]]> By: Tony Hasler http://antognini.ch/2010/01/does-the-query-optimizer-cost-px-distribution-methods/comment-page-1/#comment-7539 Tony Hasler Thu, 04 Feb 2010 11:11:41 +0000 http://antognini.ch/?p=45#comment-7539 The only stuff I have been able to find is <a href="http://download.oracle.com/docs/cd/B19306_01/server.102/b14200/sql_elements006.htm#BABCJHAF" rel="nofollow">in the documentation for the PQ_DISTRIBUTE hint</a> but it is still quite useful. I would buy Christian's book anyway :-) The only stuff I have been able to find is in the documentation for the PQ_DISTRIBUTE hint but it is still quite useful. I would buy Christian’s book anyway :-)

]]> By: Christian Antognini http://antognini.ch/2010/01/does-the-query-optimizer-cost-px-distribution-methods/comment-page-1/#comment-7536 Christian Antognini Thu, 04 Feb 2010 09:18:24 +0000 http://antognini.ch/?p=45#comment-7536 Hi Hans-Peter Sorry, but I'm also not aware of a detailed description of the distribution methods. In my book, at page 493, I added some remarks about them. http://books.google.com/books?id=b3DIkYO2gBQC&lpg=PP1&dq=troubleshooting%20oracle%20performance&pg=PA493#v=onepage&q=&f=false HTH Chris Hi Hans-Peter

Sorry, but I’m also not aware of a detailed description of the distribution methods. In my book, at page 493, I added some remarks about them.
http://books.google.com/books?id=b3DIkYO2gBQC&lpg=PP1&dq=troubleshooting%20oracle%20performance&pg=PA493#v=onepage&q=&f=false

HTH
Chris

]]> By: Hans-Peter http://antognini.ch/2010/01/does-the-query-optimizer-cost-px-distribution-methods/comment-page-1/#comment-7531 Hans-Peter Thu, 04 Feb 2010 07:28:19 +0000 http://antognini.ch/?p=45#comment-7531 Hi Christian, Can you point my to some documentation about the px-distribution methods? I searched for it but most performance tuning books do not go into detail about it. regards Hans-Peter Hi Christian,

Can you point my to some documentation about the px-distribution methods?
I searched for it but most performance tuning books do not go into detail about it.

regards Hans-Peter

]]> By: Christian Antognini http://antognini.ch/2010/01/does-the-query-optimizer-cost-px-distribution-methods/comment-page-1/#comment-7511 Christian Antognini Wed, 03 Feb 2010 14:27:47 +0000 http://antognini.ch/?p=45#comment-7511 Hi Tony Sorry, I have no real explanation for that. It just matches something that I already observed several times... i.e. partition pruning does not always match well with PWJ. To understand what's going on, you should have a look to the 10053 trace. Cheers, Chris Hi Tony

Sorry, I have no real explanation for that. It just matches something that I already observed several times… i.e. partition pruning does not always match well with PWJ. To understand what’s going on, you should have a look to the 10053 trace.

Cheers,
Chris

]]> By: Tony Hasler http://antognini.ch/2010/01/does-the-query-optimizer-cost-px-distribution-methods/comment-page-1/#comment-7507 Tony Hasler Wed, 03 Feb 2010 13:02:23 +0000 http://antognini.ch/?p=45#comment-7507 Christian, your comment reminded me of my test case. The two tables that I joined were both range partitioned by date and subpartitioned by hash on the join column. The query I ran selected a specific date in the WHERE clause. The hint was required to get the "none none" distribution mechanism despite that being demonstrably faster. Can you explain this or do I need to look at the 10053 trace? Thanks again. Christian, your comment reminded me of my test case. The two tables that I joined were both range partitioned by date and subpartitioned by hash on the join column. The query I ran selected a specific date in the WHERE clause. The hint was required to get the “none none” distribution mechanism despite that being demonstrably faster.
Can you explain this or do I need to look at the 10053 trace?

Thanks again.

]]> By: Greg Rahn http://antognini.ch/2010/01/does-the-query-optimizer-cost-px-distribution-methods/comment-page-1/#comment-7472 Greg Rahn Mon, 01 Feb 2010 22:19:21 +0000 http://antognini.ch/?p=45#comment-7472 @Tony In all likelihood your plan was not a FPWJ even though you joined on the partition keys. There would be no row redistribution (none-none) for a FPWJ plan. The DOP influences whether or not you get a FPWJ or whether there is redistribution. So the answer to your question is yes; if the DOP does not match the number of partitions needed to be joined for a FPWJ plan then you will see a row redistribution (hash-hash) plan. @Tony

In all likelihood your plan was not a FPWJ even though you joined on the partition keys. There would be no row redistribution (none-none) for a FPWJ plan. The DOP influences whether or not you get a FPWJ or whether there is redistribution.

So the answer to your question is yes; if the DOP does not match the number of partitions needed to be joined for a FPWJ plan then you will see a row redistribution (hash-hash) plan.

]]>