I would have hoped to see some benchmark latency numbers. What does it mean to easily handle a 1M clients?