HALP - Heuristic Aided Learned Preference Eviction Policy for YouTube Content Delivery Network

Explore a conference talk that delves into HALP (Heuristic Aided Learned Preference), an innovative caching algorithm designed to enhance the efficiency of YouTube's Content Delivery Network (CDN). Learn how this machine learning-based approach addresses key challenges in large-scale production environments, including computational overhead, robust byte miss ratio improvement, and impact measurement under production noise. Discover how HALP augments heuristic policies with machine learning to achieve low CPU overhead and significant performance gains. Gain insights into the algorithm's implementation in YouTube's CDN production since early 2022, resulting in an average 9.1% reduction in byte miss during peak periods while maintaining a modest 1.8% CPU overhead. Understand the importance of cache efficiency in video streaming services and how HALP contributes to providing a seamless user experience for one of the largest web applications in production.

NSDI '23 - HALP: Heuristic Aided Learned Preference Eviction Policy for YouTube Content Delivery...