LinnOS - Predictability on Unpredictable Flash Storage with a Light Neural Network
Explore a groundbreaking conference talk on LinnOS, an operating system that employs a light neural network to predict SSD performance at a per-IO granularity, enhancing predictability for parallel storage applications. Delve into the innovative approach that supports black-box devices and real production traces without user input, outperforming industrial mechanisms. Discover how LinnOS improves average I/O latencies by 9.6-79.6% with 87-97% inference accuracy and minimal 4-6μs overhead per I/O. Learn about the potential of integrating machine learning into operating systems for real-time decision-making, covering topics such as the importance of storage, current solutions, design and mitigation challenges, and comprehensive evaluation results.