Comments on: When 11g SQL is faster than 10g without any plan change… http://www.pythian.com/blogs/542/my-work-before-12g-goes-out News and views from Pythian DBAs Fri, 5 Dec 2008 03:09:11 +0000 http://wordpress.org/?v=2.6.5 By: Karl Reitschuster http://www.pythian.com/blogs/542/my-work-before-12g-goes-out#comment-87341 Karl Reitschuster Tue, 21 Aug 2007 06:19:09 +0000 http://www.pythian.com/blogs/542/my-work-before-12g-goes-out#comment-87341 Hi, in the 10g example it needed 100556 Block accesses - means to me 100000 Lookups fo the table 556 for the index scan in the 11g example the number blocks of index scans is the same : 3373 2817 aggregated (index rowid cached) lookups (some rowid processing ...) 556 for the index scans how many blocks has your table? Karl Hi,
in the 10g example it needed 100556 Block accesses - means to me
100000 Lookups fo the table
556 for the index scan

in the 11g example the number blocks of index scans is the same : 3373
2817 aggregated (index rowid cached) lookups (some rowid processing …)
556 for the index scans

how many blocks has your table?
Karl

]]> By: arkzoyd http://www.pythian.com/blogs/542/my-work-before-12g-goes-out#comment-83861 arkzoyd Sat, 11 Aug 2007 12:57:35 +0000 http://www.pythian.com/blogs/542/my-work-before-12g-goes-out#comment-83861 More tests will help figure out how it works. I don't know exactly the detail. Try to raise the index clustering factor and you'll see the number of consistent gets raising too. From what I understand, instead of managing a row at a time from the 1st step of the NL, it deals with severals (That's why it uses to manage more 100 000 blocks for 100 000 rows and now less). I've not been able to truly demonstrate it but I guess it works per block. Anyway what can been seen fom all the tests I've done, it's always more efficient than 10g from a consistent gets perspective. Gregory More tests will help figure out how it works. I don’t know exactly the detail.

Try to raise the index clustering factor and you’ll see the number of consistent gets raising too. From what I understand, instead of managing a row at a time from the 1st step of the NL, it deals with severals (That’s why it uses to manage more 100 000 blocks for 100 000 rows and now less). I’ve not been able to truly demonstrate it but I guess it works per block. Anyway what can been seen fom all the tests I’ve done, it’s always more efficient than 10g from a consistent gets perspective.

Gregory

]]> By: Amit Poddar http://www.pythian.com/blogs/542/my-work-before-12g-goes-out#comment-83631 Amit Poddar Fri, 10 Aug 2007 21:10:46 +0000 http://www.pythian.com/blogs/542/my-work-before-12g-goes-out#comment-83631 Hi, Could you help me understand what modification in the algorightm has so dramitcally brought down the consistent gets ? amit Hi,

Could you help me understand what modification in the algorightm has so dramitcally brought down the consistent gets ?

amit

]]>