Explore the innovative use of eBPF in CPU scheduling through this insightful 40-minute conference talk from the Linux Plumbers Conference. Delve into scheduling latency profiling techniques, including cgroup-oriented profiling and distribution reporting. Discover the advantages of utilizing BPF for profiling and core scheduling, and learn about measuring forced idle time. Gain an understanding of the GHOST scheduler, its messages and transactions, and various multicore scheduler styles. Examine how BPF can be employed to respond swiftly to events and how BPF programs integrate with the agent. Investigate the concept of global scheduling with BPF-PNT, address wakeup considerations, and contemplate the necessity of a userspace agent in this comprehensive exploration of advanced CPU scheduling techniques.
Overview
Syllabus
Intro
Scheduling Latency Profiling
Profile scheduling latencies
Cgroup-oriented profiling
Report as distributions
Wins by using BPF for profiling
Core scheduling
Measure forced idle time
BPF Solution
Algorithm
Take away
What is ghost?
Messages and Transactions
Various Multicore Scheduler Styles
Use BPF to respond quickly to events
BPF Programs are part of the Agent
Example: BPF scheduler with a Global Agent
Global Scheduling with BPF-PNT
What about wakeups?
Do you need a userspace agent?
Taught by
Linux Plumbers Conference
