How do you measure the impact of latency on user experience in a massive scale environment? What approaches would you take to minimize it?