The bottom line is to abandon using ndbm and use either the Berkeley db package
or GNU's gdbm. We use the latter (v 1.7.3) and it is working fine. Tim Howes
has been recommending the Berkeley db stuff (which is btree-based technology
(or hash)). gdbm is hash-based.
Tim said...
I do not believe that slapd 3.3 is taking any advantage of the
btree ordering, so in other words you should be able to do everything
with hash that you can do with btree in 3.3. But, we've done a lot
more testing with btree than with hash, so there is that advantage.
If the choice is arbitrary, I'd pick btree.
Altho, I have been doing extensive testing on gdbm and it is working quite
well. We'll probably stick with that until there's some specific reason to move
onto something else (e.g. slapd making use of specific btree-based features
like ordering).
Jeff