Improve Load Balancing With Machine Learning Techniques Based on the Sched_ext Framework

Explore how to enhance CPU scheduling performance through machine learning techniques in this 47-minute conference talk from the Linux Foundation. Learn about a novel method that leverages machine learning to extract key features for task migration, enabling dynamic and stable workload optimization in CPU imbalance scenarios. Discover how the approach builds upon the sched_ext framework, which integrates eBPF to support user-defined scheduling policies within the Linux kernel. Understand why conventional CPU utilization maximization approaches often overlook contention for lower-level hardware resources, creating performance bottlenecks in compute-intensive servers. Examine how an ML-based, resource-aware load balancer effectively addresses these imbalances while allowing data collection and model inference without kernel modifications. Review experimental results demonstrating how this ML-driven approach can outperform EEVDF for certain workloads, delivering notable performance gains for CPU schedulers.