How many concurrent connections can the service handle and how to estimate this based on 1 million daily active users, peak load, and connection lifetime?