> Lately, I have been thinking that a cool form of index would be some
sort
> of "persistent reference" index. Like the old ISAM days of yore, a
fixed
> number could point you right to the row that you want. I'm not sure if
the
> "persistent reference" is a specific auto numbering column type or
> separate index structure or both.
What you are talking about is a 'relative file'. It turns out on modern
ISAM file systems, the win you get over b-tree indexing is not worth
losing the ability to do simple things like run-length compression on
strings.
Anyways, while storing a physical offset is O(1), so is computing a
hash. How would a hash index not fill your need?
Merlin