Re: Enable 32K Block in 8K Block DB

"Mike Ault" <mikerault_at_earthlink.net> wrote in message news:37fab3ab.0403291707.6f43b737_at_posting.google.com... SNIP
> 3 - Multiple blocksizes
>
> Multiple blocksizes are used in ALL the Oracle10g TPC-C benchmarks
> because they allow far faster transaction execution with less I/O.
> These vendors spent hundreds of thousands of dollars to get the
> fastest performance and their choice of multiple blocksizes speaks for
> itself.
>
> UNISYS benchmark: 250,000 transactions per minute.
>
> http://www.tpc.org/results/FDR/TPCC/unisys_es7000-420_291K_040209_fdr.pdf
>
> db_16k_cache_size = 15010M
> db_8k_cache_size = 1024M
> db_cache_size = 8096M
> db_keep_cache_size = 78000M
>
> HP benchmark: 1,000,000 transactions per minute.
>
>

http://www.tpc.org/results/FDR/TPCC/HP%20Integrity%20rx5670%20Cluster%2064P_FDR.pdf
>
> db_cache_size = 4000M
> db_recycle_cache_size = 500M
> db_8k_cache_size = 200M
> db_16k_cache_size = 4056M
> db_2k_cache_size = 35430M
>
> Now, post some actual tests that prove or disprove these and we can
> discuss this some more. Until then, let's move on.
>
> Mike Ault

Mike,

Thank you for posting these links. I had a look at the UNISYS benchmark document.

In that document, I found the following: (1) Default blocksize is 2K as set by p_create.ora file on p.B-38 (2) Block size for the UNDO tablespace is 8K as set by the createdb.sql script on p.B-3.
(3) All other tablespaces are created with blocksize 16K as described in the createts_gen.sh script on p.B-12. This script calls the createts.sh script on p.B-10 where the blocksize is passed in as the tenth parameter.

However, this appears to contradict the blocksizes given in Appendix E where the blocksizes are given as 2K except for the following segment names: (1) ORDRCLUSTER_QUEUE 16K
(2) ROLL_SEG 8K (most likely their name for the automatic UNDO segments).

If Appendix E is more accurate than digging around in the scripts of Appendix B, then a blocksize of 2K was chosen for some reason in nearly all the cases.

On p.C-12 and forward, the bytes per sector is given as 512 for the underlying disk hardware.

On p.C-2 and C-3, the prefetch multiplier parameter is set to 4 for both used for LUN-0 and LUN-1. This means that one read operation will get 4 sectors or 2K.

It appears that this document follows (mostly) the rule that the database blocksize should match what is read in one I/O operation off the disk.

Douglas Hawthorne Received on Mon Mar 29 2004 - 20:50:35 CST