Our statsd cluster is handling 7,500 reqs/sec right now

We're currently rolling back the change that added an unruly statsd logger in our stack. As the title implies, our statsd servers are handling around half a million UDP messages per minute. I can't speak to how much RAM, etc. the servers have, but I can say that the service has held up, despite the fact that we can't handle enough I/O to send process the data fast enough for Graphite.