matthew sporleder writes: > I know this consultant situation well, and am trying to avoid it right > now. The problem really is that I have 10+ million entries, so each > write (this is a writing master. Almost never any searches) with an > index = many writes. I'll drop it down to object class and see how > performance looks from there.
You might want to set the "unchecked" sizelimit, for a max number of entries to examine _after_ indexes have been applied. That way you get instant adminLimitExceeded (and no entries) instead of a slow sizeLimitExceeded for an attempt to examine your millions of entries. If you in addition use loglevel 256 and grep the syslog output for err=11 (adminLimitExceeded), you'll discover any indexes you need to add - or maybe clients which should use different searches. Beware that what is indexed is a hash of the value, not the value itself (independently of whether the database is using a hash table) - so leave room for hash collisions. That applies even to the DN used to look up an entry. -- Hallvard
