>Tore,
>>
>>As I wrote, I NEVER use IDX, I ALWAYS use cdx. I suggest that you do the same. I see no reason to use idx, since cdx is faster AND supported during transactions. I think you should see this lack of support for IDX as a clear hint!
>
>Any data on your speed claim ? The last time I checked (HW a LOT different fromn today's) idx was (not much) smaller and faster. And I am using it only in single user datamining.
IDX can use the CDX type of indexing, if built with COMPACT clause. Then they are the same, but the header, IIRC, still shows "this is a single-tag index file".
IDX of the old type is not compressed, so it should be faster to write, at least while the B-tree is not too deep (which may hold true longer for short key expressions) and while the whole file is cached. As soon as it grows big enough, a compact index is faster - maybe slower to write, but definitely helping Rushmore achieve the speed on reads, selects, filters etc.