FetchBPF: Customizable Prefetching Policies in Linux with eBPF
summary
FetchBPF extends eBPF to let developers replace Linux’s sequential prefetcher by loading programs at hook A (collect fault addresses) and hook B (run policy) on the page-fault path, using ebpf helpers to issue fetches. The implementation reproduces VMA, swap-entry, and LEAP-style strided prefetchers, is deployed without kernel source changes, and microbenchmarks show equivalent accuracy and coverage to native kernel implementations with negligible overhead.
AI-written summary · about
chapters (7)
- 0:00FetchBPF introduction
- 3:14Kernel prefetching limits
- 3:55FetchBPF with eBPF
- 6:00FetchBPF design
- 7:27Example policy
- 7:49Evaluation setup
- 10:39Summary
related talks
47mDive into BPF performance tools using python
51mPly: Lightweight BPF Tracing
38mMy First Monitoring with eBPF
08mLightning Talk: BTFGen: one Step Closer to Truly Portable eBPF Programs
27mBeyond printf & tcpdump: Debugging Kubernetes Networking with eBPF
25mPowering Observability with Simple, Sharable, EBPF Programs
36mPanel Disucssion: Is There Actually a Byte Behind All the Buzz? eBPF in Production!
29mPostcopy Preemption
