Re: LIMIT Optimization

On Mon, Jan 28, 2002 at 04:26:44PM +0300, Oleg Bartunov wrote:
> On Sun, 27 Jan 2002, Ross J. Reedstrom wrote:
> 
> It's still a big win ! think about difference between full sorting of  1mln
> rows and partial sorting when you stop sorting after getting desirable
> first 100 (or so) rows. This is most common situation in web applications

Right - my reason for commenting was the loose language: _you_ know and I
know and Bruce knows and Tom knows what 'stop sorting after getting the first
100 rows' means, but for the sake of the archives, and TODO list, a clearer
wording might be in order.

> at least - you always display search results page by page. But statistics
> shows that 90% of hits is the first 1-2 pages. If we intend to be more
> friendly and compete with MySQL on Web apps. we should consider this
> optimization. We had quick and dirty patch for 7.1 but it was just a
> sketch and we didn't surprised core developers reject it. Now we have
> libpsort - a library which implements partial sorting, and we use it
> extensively  in our apps. I think we should add partial sorting
> in TODO list for 7.3.

Agreed it's a good idea - maybe post your 'quick & dirty' to patches
'for discussion only' as an example for people looking at the TODO
for things to try. We've had a few of those (people looking, that is)
in the last couple weeks.

Ross