Rajomon - Decentralized and Coordinated Overload Control for Latency-Sensitive Microservices

Learn about RAJOMON, a market-based overload control system designed for large microservice graphs in this 16-minute conference presentation from NSDI '25. Discover how this innovative system addresses the critical challenge of overload management in decentralized, interdependent microservices that often experience traffic surges exceeding service capacity. Explore RAJOMON's unique approach using distributed rate-limiting and load shedding, where clients attach tokens to requests and services charge prices for each API, automatically dropping requests with insufficient tokens. Understand how tokens and prices propagate through entire call graphs by piggybacking on requests and responses, making RAJOMON the first decentralized, end-to-end overload control system. Examine the implementation and evaluation results conducted on setups with up to 140 cores across various academic and industry applications, demonstrating how RAJOMON protects microservice goodput and tail latency from substantial demand spikes, even with mixed request types and deeper service graphs. Review the impressive performance improvements, including 78% reduction in tail latency and 45% increase in goodput compared to state-of-the-art overload control systems for microservices in high-load scenarios.