Hallvard B Furuseth wrote:
Gaurav Gugnani wrote:
Actually, i want to know - how to "scale out" once you reach the limits to run openLdap in one single box?
You said "some million of records". That's nowhere near OpenLDAP's limits, nor near the multi-terabyte databases you mention, unless your LDAP entries are quite large - e.g. lots of JPEG photos and the like.
Your scenario just sounds like a database which does not all fit in RAM. The Tuning section of the Admin Guide describes which parameters to give priority in that case. But as Howard mentions, that'll become unnecessary. The MDB backend will leave that to the OS.
Anyway, if you do reach those limits, I guess you must currently split up your LDAP directory. Put different subtrees in different servers. Then set up referrals between them. Tie them together with the chain overlay or ldap backend if you don't want the clients to have to deal with referrals, though that increases the server load.
Back when I wrote about that, I was speaking of back-ndb. Since it uses MySQL Cluster, you can simply add more cluster nodes if you want to scale further.
Going back to the original question - once you reach the limits of a single box, you obviously need either a larger box, or more boxes.
Unfortunately back-ndb (and the NDB API) needs a bit more work before it can be generally useful. And in the time since Oracle acquired Sun (and therefore MySQL), most people who were interested in the NDB OpenLDAP code have walked away from it. If you know of any developers who'd like to pick up back-ndb and push it further, send them over...