PeeR: First-Class Scheduling for Latency-Critical eBPF Applications
Talkat 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26) · 2026-07-13..07-15·
summary
PeeR makes eBPF programs preemptible and schedulable by inserting a lightweight two-cycle budget check at each helper call, saving interrupted state into a continuation buffer, and resuming work on per-CPU kernel threads under the scheduler (integrated with schedext). Verification-time pointer metadata is preserved to patch stack pointers on resume. Evaluations on Redis, Memcached, echo-server, and TPC-C show p99 latency drops from 3× up to 19.8× versus stock eBPF, eliminating head-of-line blocking while preserving throughput.
AI-written summary · about
chapters (15)
- 0:00Introduction to Peer and eBPF
- 1:01eBPF for key-value stores
- 2:25eBPF runtime properties
- 3:51eBPF challenges
- 4:55Intra-application contention
- 5:36Inter-application contention
- 6:59PREEMPTIBLE eBPF runtime
- 8:43PREEMPT design and scheduler
- 9:25Safe preemption points
- 10:29Pointer management
- 12:14Verification-time pointer tracking
- 13:18Peer performance evaluation
- 14:20Peer co-location evaluation
- 15:24Peer as first-class eBPF scheduling
- 15:45Conclusion
related talks
30mScheduling at Scale: eBPF Schedulers with Sched_ext
32meBPF: BPF kernel infrastructure
48mInto the hive of eBPF!
18mNoisy Neighbor Detection with eBPF
31meBPF 101
40mPanel Discussion: eBPF: A New Era in Cloud Infrastructure Tools
32meBPF Kernel Scheduling with Ghost
35mAn Introduction to eBPF - Superpowers for Networking, Observability & Security