server sizing input?

["Brian K. Jones" <jonesy@CS.Princeton.EDU>]

Hi all, 

I'm not really having a problem per se, but I'm curious about what
methodologies you all are using to come up with the right hardware specs
to use in an OpenLDAP deployment. I'll toss out some generic
assumptions, so there's some context to speak in:

Master server is OpenLDAP 2.1.22 with bdb 4.1.25 as a backend. 

The server replaces the NIS passwd, group, netgroup and ethers maps.
There are 1000 users, and 1000 hosts. 

The server will be used largely for authentication, and whitepages
service will be available to people on the local network.

So, given this information, how would you go about deciding how much
ram, how much cpu power, etc. 

Can Openldap take advantage of multiple CPUs, or is that a waste? 

At what point do you run into decreased returns with indexes? 

Is there a caching mechanism local to OpenLDAP, or is this handled by an
nscd-like system daemon? 

Anyone have links to docs regarding performance of OpenLDAP? BDB?  I saw
a good one on BDB, but can't find it now :-( I've never seen a good one
on OpenLDAP. 

In the end, what hardware are *you* using, and how many
entries/operations are you supporting? Are you happy with it?

Thanks a million for input on any of this.
brian.