From Fragmentation to Integration: Enhancing sched_ext BPF Scheduler Interoperabili...
summary
Work over a year on multiple schedulers reveals duplicated effort, user-space topology parsing, and five different ways of accessing topology across implementations. A concrete fix shows up: reuse kernel per-CPU macros so a simple function reads in-kernel topology instead of parsing user memory — it already works in one implementation. Proposed integration paths include BPF arenas, kfuncs, and scheduler governors to share data structures and APIs, plus using AI as a knowledge bridge to reduce siloed expertise.
AI-written summary · about
chapters (5)
- 0:00Introduction and motivation
- 2:32Fragmentation in BPF schedulers
- 7:29Topology challenges in Sketex
- 10:01Solutions for better interoperability
- 12:30Discussion and audience questions
related talks
30mScheduling at Scale: eBPF Schedulers with Sched_ext
32meBPF Kernel Scheduling with Ghost
34mWhat's new in the BPF world?
40mPanel Discussion: eBPF: A New Era in Cloud Infrastructure Tools
32meBPF: BPF kernel infrastructure
30mWhen BPF programs need to die: exploring the design space for early BPF...
50mIntroduction to eBPF
35mAn Introduction to eBPF - Superpowers for Networking, Observability & Security