> ... Think of a query like this:
>
> select a,b,c from table where ( a>min_a and a<max_a ) and ( b>min_b and b<max_b )
>
> In a conventional implementation you have two indexes on attributes a and b.
> But to run this query the database engine profits only from one index. It has
> to run through all the values of the other. This gets even worse if you use more
> constraints, and this scheme is typical for things like OLAP.
>
> With the new methode you add one UB-index that embraces a and b. And you run
> only once through this index.
And this is different from a multicolumn btree index how?
I looked at the referenced website when this message first went by,
and was unhappy at the apparently proprietary nature of the technology
(not to mention the excessive hype ratio). I lost interest ...
regards, tom lane