-----BEGIN PGP SIGNED MESSAGE-----
Leigh Porter wrote:
| Hiya Folks,
| OK, I have this LDAP tree running on 2.2.20 on Linux 2.6 on a AMD64
| 3200+ with about 2Gb of
| memory. There are around 3500 quite small entries, I am running a
| simple search on multiple machines
| with the following:
| for i in `cat email`
| ldapsearch -LLL -h db1 -x -b "dc=testdomain,dc=co,dc=uk"
| "(email@example.com)" mailMessageStore > /dev/null
| I run say 5 of these on about 3 boxes just to give the LDAP box some
I did a very similar test (I guess you're also benchmarking for
qmail-ldap) a few months ago, with OpenLDAP-2.1.25->2.1.30 on bdb
backend with db-220.127.116.11, on Dell PowerEdge 1750's (IIRC 2.8GHz Xeon's)
with 2GB ram, with 250 000 entries in the directory. I ran 4 instances
of my test script on each of 3 machines, and consistently achieved 200
queries/sec. OpenLDAP-2.2.x should do quite a bit better ...
| The problem is that no matter how many instances of this test script I
| run, I only ever get about 50 searches
| per second form the LDAP server, I would have expected more..
| Thanks to the helpful comments from this list I have created a
| with the following:
| # Set the database in memory cache size - get the whole thing in
| set_cachesize 0 102400000 0
| # Set database flags.
| set_flags DB_TXN_NOSYNC
You should not do this except for mass loading via slapadd.
| # Set log values.
| set_lg_regionmax 1048576
| set_lg_max 10485760
| set_lg_bsize 2097152
| set_lg_dir /usr/local/var/openldap-data/logs
| # Set temporary file creation directory.
| set_tmp_dir /tmp
| The large cache is to get the whole database into memory.
| Here is slapd.conf (relavent bits only)
| threads 100
You shouldn't need that many. You might be over-tuning here (+- 8MB
memory required per thread IIRC). IIRC I was running with 36.
| loglevel 0
| database bdb
| index objectClass eq
| index mailMessageStore sub,eq,pres
^ Do you really have clients searching on the mailMessageStore that
| index mailServices sub,eq,pres
Your searches use a searchfilter of "mail=<string>", yet you aren't
indexing mail. You probably need something like:
index mail eq,pres
| sizelimit unlimited
Bad idea ...
| cachesize 10000000
Caching 10 000 000 LDAP entries when you only have 3500 seems a bit
| dbcachesize 10000000
This one is deprecated by the BDB configuration in DB_CONFIG
| This is what I get from checking the Search Operations Initiated on the
| monitor db:
| (never more than 50 searches/s if that)
| monitorOpInitiated: 4000
| monitorOpInitiated: 4044
| monitorOpInitiated: 4088
| monitorOpInitiated: 4132
| monitorOpInitiated: 4176
| monitorOpInitiated: 4220
| monitorOpInitiated: 4264
| monitorOpInitiated: 4309
| monitorOpInitiated: 4354
| monitorOpInitiated: 4396
| monitorOpInitiated: 4439
| monitorOpInitiated: 4484
| Assuming my attempt at a DB_CONFIG and the slapd config are OK, I am
| thinking that this may be a limitation
| somewhere els eon the box.. Can anybody help out with this please?
Index the attributes you use in your search filters ...
Buchan Milne Senior Support Technician
Obsidian Systems http://www.obsidian.co.za
B.Eng RHCE (803004789010797)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org
-----END PGP SIGNATURE-----