question about caching policy

["Kwon, YongChul" <godslord@sparcs.kaist.ac.kr>]

Hello. I have a question in OpenLDAP caching policy.

1. There is a backend object. and backend object has
10 buckets of entry(keeps in memory). if a search query
issued by client, slapd looks up the 10 buckets. if fails,
it open ldbm and fetch the requested entries. and update 
cache. then, slapd keeps only 10 entries in whole connection?
I want to know that the cache is managed by a session or 
the entire server process. if it compiled with multi thread,
does the caching policy changed?

2. What's the ldbmcachesize means? in gdbm, it allocates memory
of cachesize - it keeps until the gdbm session closed. then,
the ldbmcache has no meaning in slapd with gdbm. it closed the
gdbm session when search operation ended. am I correct? :-(

How can I use the best perfomance of OpenLDAP - with gdbm or
Berkeley DB 2.x?

please answer soon. 

Thank you.