The poll data might be the output from Apache's mod_status or we might enable remote uptime checks (such as rup (1)) or perhaps write our own server status daemon.

The efficacy of such an approach hinges on short TTL's. For instance, if our polling interval is every 300 seconds, our TTL's should be no longer than that. Expirations need to be short as well so that secondary DNS servers will stay in sync.

The extra work that the DNS servers are performing may be insulated from our domain's zone by defining a seperate zone, say, dynamic.example.com and we can have www.example.com simply CNAME'd to www.dynamic.example.com -- DNS for the dynamic zone can be a box beefed up enough to handle all the extra effort.

Ideally, 60 second TTL's and poll intervals leave us with shortest failure recovery time and highest load balancing granularity.