Distributing Global State to Serve over 1 Billion Daily Requests

Render has evolved its networking layer to efficiently handle over one billion daily requests across multiple regions by optimizing its state distribution and caching mechanisms. Initially, Render operated in a single region with a basic in-memory caching system that required frequent requests to a central service database, leading to limitations such as cache wipes and database overloads during redeployments. To address these issues, Render transitioned to a push-based caching system where routers receive updates directly from the service database, reducing latency and improving update propagation. As Render expanded to additional regions like Frankfurt, Ohio, and Singapore, they implemented shared cache instances within each cluster to decouple routers from the service database, thereby enhancing startup times and maintaining scalability. This architectural evolution has allowed Render to maintain low-latency routing and high availability, even during significant traffic spikes, while preparing the platform to support further global expansion and increased user demand.