RL-Watchdog: A Fast and Predictable SSD Liveness Watchdog on Storage Systems

Explore a groundbreaking conference talk on RL-Watchdog, a novel reinforcement learning-based watchdog system for monitoring solid-state drive (SSD) liveness and detecting failures. Discover how this innovative approach utilizes a lightweight watchdog, reinforcement learning-based timeout prediction, and fast failure notification to minimize application data loss in storage systems. Learn about the implementation of RL-Watchdog in Linux kernel 6.0.0 and its impressive performance in reducing data loss by up to 96.7% compared to existing schemes, with failure point prediction accuracy reaching 99.8%. Gain insights into the potential impact of this technology on improving SSD reliability and data protection in modern storage systems.

USENIX ATC '24 - RL-Watchdog: A Fast and Predictable SSD Liveness Watchdog on Storage Systems