Cloud-load testing service is hyper-scale ready: lessons from generating 1M concurrent user load

This post details infrastructure changes to the Cloud Load Testing (CLT) service that enable it to generate up to 1 million concurrent user load. It highlights improvements in agent efficiency (2.5x more virtual users per core) and reduced resource acquisition time (over 50%). The post also delves into test settings, including the 'webtest connection model' (Connection Per User vs. Connection Pool) and provides a formula for calculating connection pool size based on dependent requests and server URI hosts. It discusses agent capacity planning and the use of performance counters like TCPv4 for hyper-scale testing.