[Bizgres-general] A bitmap index access method is available in bizgres CVS tip

[Mark Kirkwood]

Simon Riggs wrote:

> 
> Real data is seldom uniformly randomly distributed when there are more
> than a few values. The most frequently occurring natural distribution is
> Zipf. With that, as we add more distinct values the frequency of
> occurrence of those additional values reduces and therefore the
> compressibility of the additional bitmaps increases also.


> Think about
> how big the bitmap for a value would be if only 1 row had that value.
> 

Err - for a conventional bitmap it's always the same, so matter how many 
rows actually have the value (e.g. for 10e7 rows like the example):

bitmap for 1 value size = 10000000/8 ~ 1.1Mb

Now clearly the confusion here is because I have not really got my head 
around the Bizgres bitmap implementation yet (tho, Jie has provided some 
pointers, and I'm reading the code today). If you have a nice 
description of the algorithm you used, then that would be much appreciated!

Cheers

Mark