Index substring min length 3?

[Rich Graves <rcgraves@brandeis.edu>]

We have finally entered the 21st century by upgrading from OpenLDAP 
1.2.13 to 2.0.27. Platform is RedHat 7.3, minimal changes to the latest 
errata they've packaged. ldbm backend, gdbm libs.

Problem:

Under openldap 1.2.13, we were able to get indexed searches on 
3-character substrings with acceptable performance. For example,

ldapsearch 'cn=*esz*'

would return the 4 matching people instantly. Under openldap 2.0.27, the 
same search takes about 10 seconds.

Index specifications are the same in either case,

index cn pres,eq,sub

The cutoff seems to be 3 characters. 4-character search terms, and 
initial/final searches (presumably the boundary is a 4th character), are 
speedy as ever.

In servers/slapd/slap.h I have found

#define SLAP_INDEX_SUBSTR_MINLEN        2
#define SLAP_INDEX_SUBSTR_MAXLEN        4

but my limited understanding doesn't read that as a magic number of 3 
characters for a full table scan. Do I want to fiddle these defines, or 
am I on the wrong track.

I have mitigated the impact by rewriting our web/finger interfaces to 
change the filter specs from cn=*%s%* to exact sn or initial/final 
anchored cn searches, but I'd prefer to understand where the problem is, 
and either fix it or be talked out of allowing 3-character substring 
indexes because it's a silly thing to do.

We only have a total 9000 directory entries, so our indexes are not 
outrageously large. Tuningwise, I've settled on

dbsync          10
cachesize       2000
dbcachesize     1500000

which gives us a RSS of about 100MB.
--
Rich Graves <rcgraves@brandeis.edu>
UNet Systems Administrator