From the free ipa mailing list:

> In the beginning of the fedora 15 user document there is comment on load balancing yet when you join a client its stating a specific server, so how does that work?
> 

Clients should use SRV DNS records for load balancing and IIRC they are
configured to do so automatically.

> So is the load balancing manual-matic?  ie if you have several replicas  you have to point the client at one or other when you join?
> 

No, it's automatic.

This is how it works in detail:
There are two important parts of a SRV records - priority and weight.
Priority specifies the order in which the servers should be contacted.
The client first attempts to contact the server with the lowest
priority, if it doesn't respond, then try the second lowest etc.

Weight specifies the relative load for entries with the same priority.

All replicas are assigned the same priority by default. Check out "ipa
help dns" for some srv examples if you need to change the proiority.

> If so how does the client survive while one replica is off line?

The client includes a pretty sophisticated fail over mechanism that
seamlessly reconnects to another configured server if the current one does
not respond