Re: 10046 trace: 'latch free' then 'db file seq. read' waits

Spendius wrote:
> joel garry a écrit:
> >
> > Conversely, maybe you have the same number of very hot blocks that are
> > the last few that the scan needs, and everybody is fighting to get them
> > to see the data in their own consistent way?
>
> That's what I guess too of course, but
> o I'd like to be able to verify it,
> o and answer this question: why are these last records to be displayed
> *always* located in these assumingly hot blocks ?

You will have to understand your application. Are the last records being displayed the most recent or least recent? Are these the ones that are getting concurrent online activity?

Did you understand the comments noted earlier about your running FIRST_ROWS and the optimizer being biased to get "some rows quickly" versus all the rows in how it creates an execution plan?

I think you have been given some good hints ... now go forth and solve it!

But get a complete 10046 trace and look at it with a decent resource profiler. Received on Fri Jun 16 2006 - 15:29:50 CDT