Re: Histogram worthwhile?
Date: Tue, 24 Feb 2009 22:30:32 +0800
I have two Blog Posts on Bind Variable Peeking and Histograms :
http://hemantoracledba.blogspot.com/2007/09/more-on-bind-variable-peeking-and.html (pointing to Tom Kyte's articles)
Hemant K Chitale
At 10:01 PM Tuesday, Roberto Veiga wrote:
>We have another performance problem to solve. The performance that
>was good, now is bad. Very bad with an unacceptable performance and
>nobody makes any changes to the system. Nor the analyst neither the
>dba. Why the performance is not ok anymore?
>I dont know if this is a problem or if it is just a optimizer's
>behavior, but I am doing a research to find some alternatives to
>this problem. And I find some (thanks Tom Kyte):
>.use bind variables but dont use histograms. so the plan is going to
>be the same for every single execution. one size fits all approach
>.dont use bind variables but use histograms. so the optimizer will
>evaluate the plan every execution and discover the best plan
>.use bind variables, use histogram but change the code. if the value
>is in this range, execute this query. if not, execute another query.
>i need to know distribution data. using the width_bucket function
>could be a good way to discover how is the distribution
>Anyone have another idea?