Multiple DB DBLOCK SIZE for warehouse database?Suggest

Question

Hi All,a faculty from Oracle University told me the idea behind having the multple DB DBlock sizes in Oracle is not the performance but just for the Transporting tablespaces ....Just confirm is that true??????
Database architects here are using larger Block sizes for tablespaces for Warehouse database....which I guess is of not any use...what you say?
Thanx in Advance
Gagan

Answer 1

gaganahuja wrote:
Hi All,a faculty from Oracle University told me the idea behind having the multple DB DBlock sizes in Oracle is not the performance but just for the Transporting tablespaces ....Just confirm is that true??????
Database architects here are using larger Block sizes for tablespaces for Warehouse database....which I guess is of not any use...what you say?Gagan,
one important point that is often overlooked in this discussion is that the cost based optimizer does a questionable job when dealing with objects in non-standard block size tablespaces when it comes to costing full segment scans (full table scans / index fast full scans).
The Oracle runtime engine adjusts the multi-block read size in non-standard block size tablespaces accordingly, which means that e.g. given a multi-block read count (MBRC) of 8 and a default block size of 8K it will attempt to request 64K (whenever possible, there are cases where this is not possible, reaching extent boundaries or blocks already in the buffer cache).
So in a 16K non-standard block size tablespace the MBRC will be adjusted to 4 at runtime (4*16K = 64K), and the same happens e.g. in a 2K non-standard tablespace where it will be adjusted to 32 (32*2K = 64K).
So apart from effects outlined above that might lead to a reduction of the actually requested multi-block read size there is no difference at runtime with regards to performing a full segment scan.
But unfortunately the cost based optimizer doesn't reflect this runtime behaviour properly, especially if CPU costing is enabled, which is the default in 10g and later.
I've recently run some test cases regarding this particular issue which were provided by Jonathan Lewis as part of the "Cost Based Oracle - Fundamentals" book against the 10.2.0.4 and 11.1.0.6/7 release and they show that the cost estimates are at least questionable. By the way, although I've run the tests against 10.2.0.4 and 11.1.0.6/7, they correspond to the results of previous releases, so the same applies to 10.1 and 9i as well.
For more information, see here: http://oracle-randolf.blogspot.com/2009/02/cost-based-oracle-fundamentals-test.html
Regarding thoughts about putting indexes in larger block sizes, Richard Foote has some interesting notes that try to explain why it probably won't matter in most cases: http://en.wordpress.com/tag/index-block-size/
Regards,
Randolf
Oracle related stuff blog:
http://oracle-randolf.blogspot.com/
SQLTools++ for Oracle (Open source Oracle GUI for Windows):
http://www.sqltools-plusplus.org:7676/
http://sourceforge.net/projects/sqlt-pp/