Explore a 32-minute conference talk that delves into improving CPU utilization through the design of a new scheduling algorithm. Learn how large Internet companies tackle the challenge of low CPU utilization by deploying both offline CPU-bound loads and online services on a single node. Discover the innovative approach of introducing an offline scheduling class based on the Linux kernel to address this issue. Gain insights into latency sensitivity, classical problems, and proposed solutions. Examine load balancing techniques, scheduling mechanisms, generic interfaces, and off-balancing strategies. Conclude with a comprehensive understanding of how this new algorithm can optimize CPU performance and enhance the quality of online services.
Overview
Syllabus
Intro
Latency Sensitivity
Classical Issue
Solution
Load Balancing
Scheduling
Generic Interface
Off Balancing
Conclusion
Taught by
Linux Foundation