At 04:43 AM 4/6/00 +0800, Joseph Alba wrote:

>There is a selectivity issue because IB uses only one kind of index
>structure isn't it - sparse (which will have this kind of Achilles heel
>whenever the key it indexes on is not unique and are too many to traverse
>for comfort).
>
>And there is only one kind of index because IB had that label of being
>embedded, so it had to be kept small.
>
>Would it not be nicer to have not only one but maybe two index structures?
>
>On keys that are pretty unique, the designer can use the original index
>structure. (sparse)
>But for keys that have too many duplicates - thus designer can choose the
>alternative (new) index structure (dense).

Actually, the single type of index is more a system management
issue than a size issue. I have looked at a number of different
index structures, and will again. The open question is not
whether a dense index can be built - of course it can - but
whether that dense index can be used effectively.

Ann