Slavek,

>My teoretical think:
>Small linear index for fast acces (probably slow insert )
>It is useful for small tables (aprox. 10 000 records, short PK like smallint).
>This table is usualy static but much reading.
>Acces to index by split interval algorithm.
>Index will be contain sorted values and data pointers.

Wouldn't it be faster to build an array of data-pointers
indexed by the key value? Works only on tables up to
10,000 records, requires reading in the whole index
each time, doesn't fit with the page model of the
database (headers do interfere with direct access), doesn't
work (at all) with the multi-generational aspect of the
database, but ...

This is going to sound pretty harsh, but that is exactly the
sort of idea that creates monsters like Oracle where you need
someone with an advanced degree in Oracle administration to
keep the database running. The database designer must choose
between a "fast" index and one that can handle a large table.

Regards,


Ann