SKQ: Event Scheduling for Optimizing Tail Latency in a Traditional OS Kernel

on YouTube ↗15m

" summary

Schedulable Kqueue (SKQ) replaces FreeBSD kqueue internals with a scalable architecture and event scheduler that routes events to per-thread queues and implements policies like CPU affinity, queue affinity, best-of-two and a hybrid policy. Evaluation on memcached and RocksDB shows concrete wins: up to 1022× tail-latency reduction and 27.4× longer low-latency throughput in RocksDB, 33% lower tail latency in memcached, and SKQ closes 83.7% of the gap to kernel-bypass (Shenango) on an imbalanced workload.

AI-written summary · about

" chapters (20)

1. 0:00Intro
2. 0:11SKQ: Event Scheduling for Optimizing Tail Latency in a Traditional OS Kernel
3. 0:28The Latency Problem
4. 1:11Sources of Latency
5. 2:25Event-Driven Programming Models
6. 4:28SKQ Usage
7. 5:04SKQ Architecture
8. 5:33Scalability Showdown: SKQ vs. Kqueue
9. 6:18Performance Improvement: SKQ vs. Kqueue
10. 6:55Scheduling Policies
11. 7:05Challenges of Efficient Event Scheduling
12. 8:28Cache Locality Policies in Memcached
13. 9:06CPU Affinity vs. Queue Affinity
14. 9:37Imbalanced Workload: RocksDB
15. 9:58Load Balancing Policies
16. 11:14Hybrid Policies
17. 12:04SKQ vs. Kernel Bypass: Uniform 10 ps Workload
18. 13:06SKQ vs. Kernel Bypass: Zipf-like Workload
19. 14:31Policy Selection Guidelines
20. 14:43Conclusion