At 02:18 PM 6/10/03 +0000, Roman Rokytskyy wrote:

>Arno,
>
> > The question is, is it intresting to add a new index-structure which
> > could be used when a user want it. I think it isn't bad to have 2
> > index-structures inside a engine, but let's wait for some more
> > comments/ideas.

It would make a lot of sense to add architectural support for multiple
index schemes if for no other reason than it allows rolling upgrades of
incompatible structures. It would also make it easy to generate
testbeds to compare alternatives on an apples to apples basis.

>For example, I am working with geo data and b-tree is not very usefull
>there (simple quad-tree or r-tree give me much more better
>performance). Full-text search hardly can be implemented with b-tree
>too (inverted file structure give much more better performance). But
>these ideas belong to what usually is called "object-relational dbms".

Netfrastructure has B-Tree based full text search and it works superbly
well. When you do an IBPhoenix.com knowledge-base search, you're
using a B-Tree based full text search.



Jim Starkey