Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn advanced CPU scheduling techniques for optimizing performance on AMD chiplet architectures through this 19-minute conference talk from the Linux Plumbers Conference. Explore how Google engineers tackle the complex challenge of thread placement on machines with intricate cache hierarchies, specifically focusing on AMD CPU Core Complexes (CCXes) that are too small for traditional hard partitioning strategies. Discover an innovative extensible scheduling solution that combines BPF (Berkeley Packet Filter) technology with a userspace agent to implement per-CCX runqueues and asynchronous bin-packing algorithms for dynamic thread group management. Examine the sophisticated policy framework that maintains soft affinity while intelligently managing thread spillover during load spikes, with particular emphasis on virtual machine scheduling applications. Understand the architecture's approach to balancing fairness among competing threads while preserving high throughput and cache locality through tunable parameters and complex accounting mechanisms. Gain insights into the BPF-userspace interaction design and explore community discussions on best practices for policy tuning and potential design improvements for high-performance chiplet scheduling.
