Well, you may be right. It's been awhile since that article. I think the main thing is that they try to load the entire index up into memory. Now, did the article say that they have somehow compacted it? I can't remember, but maybe I'll find the time (ha!) to locate that article. The point was that they may have to go to disk for the actual record, but want to avoid going to disk for index data.

>There is no irrefutable documentation. :) The bitmap theory was proposed long ago and seems sensible. Holding raw index tag data would take too much memory.
>
>I didn't mean they do it just for purposes of the CONTINUE command. It also applies to SET FILTER. How can it skip to the next filtered record without some internal references? How can it COUNT a filtered record set without some internal references?