[Date Prev][Date Next]
Re: tweaks for speeding up ldap
When you see the drop in performance, are you running single
queries against your ldap instance? Or large numbers of queries
using multiple indices?
Also, turn up the loglevel in slapd.conf to see if you can determine
the source of the delay.
Are you using complex acls?
I'd guess that your performance drop happens when your indices
run out of memory.
What's the size of your index files when the performance drops?
Mine are in /var/lib/ldap .
Are you doing any swapping when the performance drops?
Watch vmstat as you add entries.
You could determine precisely which attributes you need to index
and then remove other indices.
For your situation, I'd look at three areas of optimization:
1) openldap configs
2) hardware config
1) adjust your dbcachesize setting to try to get each index
completely in memory.
How complex is your schema? Could it be segmented? If it's
just a telephone book app, then you likely won't be able
to segment much.
2) Wrt hardware, you can always get a faster processor, more memory,
more and faster disks, etc.
3) If you need to support large numbers of queries, you could load
balance the read traffic across numerous slaves of a single master.
I'm curious to hear more about your results. My directory is
much smaller ;) so I've never tested with numbers like what
you're talking about.
On Sat, Dec 14, 2002 at 03:05:07AM +0530, parthasarathi s a wrote:
> I actually found out in a search of several mailing lists that by adding dbnosync and dbnolocking to slapd.conf would show some improvement. But i didnt notice anything really very drastic. Will stripping down any code help? If it does, which part of the code should i fiddle around with. Or (i repeat) are there any compiler flags so that i can do a recompile for my machine[P3-800, running redhat linux 7.2]. I dont require any form of authentication in the server. Right now able to do around 80k-100k entries(a telephone directory) without sweat. But after that thers a drastic fall. [Am using open-ldap 2.0.7].
> I'm targetting ~10 million entries, with substabtial speed.
> Now since clustering is ruled out (almost) Any suggestions would be of great help.
Ari Jort * email@example.com
B96D F190 7661 764E F454 2F23 9D78 1C2D A11D 4B36