Re: Functions and Relations

vc wrote:
> Aloha Kakuikanu wrote:
> > vc wrote:
> > > In a sort of way, yes, it does if you think that a 5 times difference
> > > in the execution time is OK.
> >
> > Since, the difference the other way is routinely orders of magnitude, 5
> > times time difference is nothing.
> >
>
> That is odd. I've never seen "the difference the other way" being
> "orders of magnitude". There's always the first time as they say.
> Please provide a test case of a HJ being at least 10 times slower than
> a NL join for the same collection of data of course.

Table A - 1 row
Table B - 1G rows

join key is a unique index on B

NL join executioon scans table A.- 1 block read, then navigates 3-4 levels in the index which amounts to 3-4 block reads,
then access a row in table B - 1 more block read.

HJ had to scan full table B.

> But it's rather boring a subject for the theory group would
> not you agree ?

Ceased and desisted. Received on Wed Nov 22 2006 - 07:53:32 CET