Dmitry -

Perhaps I read your message too quickly.

>For example, if we have table with 10K records and 10 different values for
>an index, there will be ~1000 equal keys of each value in index.
>If I will delete 1000 records, than there will be
>~ (1000 * 1000) = 1,000,000 cycles.

That computation is pretty crude. It assumes that the index
duplicate you want will always be on the end and ignores the
fact that the duplicate chains are getting slower.

The other factor is the cache size. Whipping through a chain
of duplicates in memory isn't all that bad. Churning the cache
is a mess.


Regards,

Ann
www.ibphoenix.com
We have answers.