substring index oddity

["John Madden" <jmadden@ivytech.edu>]

This is something of a repeat of a previous post:

http://www.openldap.org/lists/openldap-software/200101/msg00542.html

I think I'm seeing the same sort of behavior.  Basically, I've got a directory
with 1 million objects on a machine with 4GB RAM, a cachesize of 100000, and a bdb
cache of 2GB.  Searches on an indexed attribute (uid, let's say) returns odd
problems off of substring searches with "index uid sub,eq,pres:"

uid=*222* : 29 seconds
uid=*222 : 0.063 seconds
uid=*2*22 : 0.14 seconds

All searches return roughly the same number of entries, give or take a couple
thousand.  All of the "good" searches take less than a second, all of the "bad"
ones take about 29 seconds every time, leading me to believe the index is simply
not being used and that's just how long it takes to traverse the entire directory.

One thing that does seem suspect is that db_stat is reporting a very low cache hit
on the bdb for the uid index, but everything appears to be in-memory and there's
absolutely zero I/O while these searches run.  (Memory was seeded before testing
with an objectclass=* search, which bumps slapd's RSS up to 1.8GB.

This is all on Debian 3.1's (amd64) libdb4.2 and slapd packages.

Any ideas here?

Thanks,
  John




-- 
John Madden
UNIX Systems Engineer
Ivy Tech Community College of Indiana
jmadden@ivytech.edu