Re: suggestion for more efficient inner join algorithm

[Paul Cadach]

Hi,

----- Original Message -----
From: "Dennis Haney" <davh@stripped>
To: "Tim Samshuijzen" <tim@stripped>
Cc: <internals@stripped>
Sent: Thursday, April 03, 2003 2:49 PM
Subject: Re: suggestion for more efficient inner join algorithm

[skipped]

> A "hack"; you can "reverse" the date you put in.
> eg. instead of
> insert into tbl whn=NOW();
> and select ... from tbl sort by whn desc;
> you do
> insert into tbl whn = NOW(), whenidx= UNIX_TIMESTAMP("2036-01-01
> 00:00:00") - UNIX_TIMESTAMP();
> and select ... from tbl sort by whenidx
> where whenidx is an int(11)

It's just a hack, real hack. If Tim's table contains about 5 millions records,
additional field enlarges data file for 4*5000000 ~= 20 Mb, and I don't know how
it will enlarge index file, but it will.

May be better is to realize reverse index walk for full descending sorts
(selects like ... order by a, b desc can't be optimized IMHO)?


WBR,
Paul.