Explore a conference talk that delves into Protego, an innovative system designed to address overload control in applications with unpredictable lock contention. Learn how this solution tackles the challenges of concurrent datacenter applications, where synchronization for shared data access can negatively impact throughput and tail latency. Discover Protego's two key contributions: a novel admission control strategy that prevents compute congestion caused by lock contention, and a latency-aware synchronization abstraction called Active Synchronization Queue Management (ASQM). Understand how these approaches work together to regulate incoming request rates and allow applications to abort requests exceeding latency objectives. Examine the practical applications of Protego in real-world scenarios, including its implementation in Lucene and Memcached, and explore the significant improvements in goodput and latency compared to existing overload control systems.
Overview
Syllabus
NSDI '23 - Protego: Overload Control for Applications with Unpredictable Lock Contention
Taught by
USENIX