>>To my best knowledge, this isn't correct. Rushmore only loads matching indexnodes into memory and searches the first one in a simular way as the SEEK command. Non-mathing indexnodes (besides the ones to reach the matching ones) are left alone and are never addressed.<<

We'll never know until Microsoft explains exactly what happens. I thought Chris Probst's infamous FPA article proved that the amount of data moved across the wire for the relatively few deleted records was large enough to account for the entire tag.

>>As you have been pointing out LOCATE is designed to work with sets of records, and has to download all the matching indexnodes into memory. Besides the other overhead of using rushmore, this is the reason that seek always will be faster than locate.<<

What other overhead of using Rushmore. Rushmore is the overhead (relative to the seek).

Mike