Re: *****SPAM***** Re: I/O and db_file_multiblock_read_count
- From: "Stefan Knecht" <knecht.stefan@xxxxxxxxx>
- To: "Wolfgang Breitling" <breitliw@xxxxxxxxxxxxx>
- Date: Sun, 10 Dec 2006 09:53:38 +0100
Ahh, you're right. I must've misinterpreted that somewhere.
Thanks for the update!
Stefan
On 12/9/06, Wolfgang Breitling <breitliw@xxxxxxxxxxxxx> wrote:
At 01:44 AM 12/9/2006, Stefan Knecht wrote:
>The thing is, when you collect system statistics (aka make use of
>cpu costing) - Oracle gathers stats about your systems' average
>multiblock read count. It then uses that information instead of
>whatever you might have set in your init.ora.
>
>Therefore, his test may be wrong - IF he's got sysstats.
That is not correct. Kevin's test is independent of whether he has
system statistics or not. System statistics come into play only
during the CBO's parse phase. If system statistics are in place then
the CBO during parsing will use the MBRC value of the system
statistics to calculate the cost of full scans, but Oracle's
execution engine will use the db_file_multiblock_read_count value
when doing full scans. That is one of the beauties of system
statistics - you can use the OS' full IO bandwidth without messing up
the CBO's cost calculation for full scans.
Regards
Wolfgang Breitling
Centrex Consulting Corporation
www.centrexcc.com
Other related posts:
