On Tue, 7 Feb 2012, Brian Reichert wrote:
The thread overall suggests the tried-and-true tactic of using slapcat to extract and LDIF file, to be imported later. But, our application's DB if large enough that reimportation is prohibitive.
First off, I'd try the latest-greatest RE24 with the latest-greatest back-mdb, and see if your slapadd performance still doesn't meet expectations. Things have come a long way from 2.3.43. A traditional dump-and-restore may not be the most fascinating option, but it's really easy to understand and it's really reliable...which may make it the best choice.
Anyway, if you're operating at this sort of insanely large scale, I'm going to enter in an assumption that you have lots of hardware, sites, etc. to go around. If that's the case, I'd take a look at the active-standby mirror mode (see OpenLDAP 2.4 Administrator's Guide, picture under section 18.3.4.1.1). That'll handle two sites that should be ready to go at any time.
If that's not paranoid enough, take n of your replica pool, and place it in additional sites (with or without live load, your call). If, let's just say, Site A is confirmed dead for a while, you could reconfigure that replica as Site A'. (In theory, at least. Don't ask me, I've never tried it, I don't have enough servers/glass/load balancers to make this happen.) If that works, you should be able to keep roaming around different data centers forever.
(Of course, you probably have other services too. So whether or not it's worth doing this with OpenLDAP Software, or if you should be looking at a lower level like virtualization, would be an interesting question in my mind...)