Hi Sergey,

>That's not true. Also, with 50% or more deleted records even SEEK will take >forever.
AFAIK Seek should take on the average only a bit longer,
since it directly walks the index, checks the record until
a record not deleted is found. If there are about 50% deleted
records (which are distributed randomly) it shouldn't make
too much a difference.

>As any other index. There's nothing special aboud index on
>DELETED() in that regard.

If there is an index deleted(), every delete marks the .cdx as
changed and any previously buffered indexes have to be discarded and read
again. If the .cdx is very large (I am thinking on levels upward
of 50MB) this will create a form of thrashing,
since any following rushmored call needs to load the .cdx again,
since the deleted() made it "dirty" on the delete.

Speaking from expiriences with data mining tables,
total volume about 60GB, split several ways:It depends
(as always) on the structure of your data and your app...

my 0.02 EUR

thomas