I built a 3.75-million subscriber chat system in an afternoon

Momento Topics, a serverless pub/sub service, was tested for its scalability by simulating a 3.75 million subscriber chat system in an afternoon, achieving consistent and impressive performance metrics. The test involved modeling 75,000 game room chats, each broadcasting messages to 50 subscribers, ultimately maintaining a message receive rate of 575,000 messages per second. The test used four c7gn.4xlarge EC2 instances and the rpc-perf load generator to optimize costs while scaling up the load. The service demonstrated consistent performance with p99.9 client latency around 51ms and subscriber end-to-end latency at ~220ms, validating its capability to support high-demand real-time applications, particularly in the gaming industry where large and variable waves of traffic are common.