Re: cross column correlation revisted

On Thu, Jul 15, 2010 at 12:04:21PM +0200, Hans-Jürgen Schönig wrote:
> hello ...
>
> a view is already nice but i think it is still too narrow.
> the problem is: you don't want a view for every potential join.
> in addition to that - ideally there is not much left of a view when it comes to checking for costs.
> so, i think, this is not the kind of approach leading to total success here.

The prolem is a very big one, and it's helpful to try solving it one piece at
a time, so the single table and view-based cases are probably good starting
points.

> one side question: does anybody happen to know how this is one in oracle or db2?

Neither appear to handle multi-column statistics in any form.

[1] http://download.oracle.com/docs/cd/B13789_01/appdev.101/b10802/d_stats.htm
[2]
http://www.ibm.com/developerworks/data/library/techarticle/dm-0606fechner/index.html

--
Joshua Tolley / eggyknap
End Point Corporation
http://www.endpoint.com