Harmonizing Efficiency and Practicability: Optimizing Resource Utilization in Serverless Computing

Explore a 21-minute conference talk from USENIX ATC '24 that delves into optimizing resource utilization in serverless computing. Learn about Jiagu, a novel approach that addresses the challenges faced by current serverless platforms in balancing efficiency and practicability. Discover two innovative techniques: pre-decision scheduling, which achieves accurate prediction without overhead by decoupling prediction and scheduling, and dual-staged scaling, which enables frequent instance adjustment with minimal overhead. Examine the implementation and evaluation of Jiagu using real-world applications and traces from public cloud platforms. Gain insights into how this approach significantly improves deployment density, reduces scheduling costs, and decreases cold start latency compared to existing QoS-aware schedulers and commercial cloud solutions.