more details about indexing

[Piotr Wadas <pwadas@jewish.org.pl>]

Hello list, 
Using 2.1.29/bdb 4.52/i386 (debian).

I've made some tweaks with indexing, setting some
attributes I want to be indexed, then I stopped
slapd, runned slapindex, and started again, as described
in docs. 
Everything looks to be working fine, I even added
some SUBSTR parameter to attributes I want to be
indexed with sub (except pres and eq), and all looks
OK.
However my question is how indexing works? There's
no more details about it in admin guide at openldap.org.
What I mean especially is where and when indexing is done?
I guess I shouldn't add to cron sth like 
"stop slapd && slapindex && start slapd" for each night,
to have all data I need indexed. So how this is working?
New data I put into directory is indexed when being put,
or what? I don't think that good idea is restarting slapd each time
I want new data to be also indexed.., guess so there's
some other ways to do it - is it maybe some application
to index slapd online (?) or some other way? What indexing
really does to provide things it is intended for (faster searching etc.)?
How indexing is related to replicas? Replicated trees should
be separately indexed on slave machines or not? 
I also guess that (from docs) "mode" and "dbcachesize" settings
are related to ldbm, while I use bdb, so I don't need to play
with them - does it mean bdb doesn't need such options, or they're
not implemented yet?
Regards,
Piotr