Self-Clocked Round-Robin Packet Scheduling
summary
Shows a new scheduler called SCRR (Self-Clocked Round-Robin) with a parameter-less algorithm that replaces DRR’s quantum with self-clocking, a Linux implementation on a physical testbed and theoretical bounds. SCRR keeps DRR fairness while reducing CPU by 23% versus DRR with a small quantum and improving application latency by 71% versus DRR with a large quantum; microbenchmarks show 4–8× lower latency and much lower CPU than poorly tuned DRR.
AI-written summary · about
chapters (11)
- 0:00Fair packet scheduling intro
- 1:05Current packet scheduler challenges
- 2:51Limitations of Deficit Round Robin
- 4:39DRR quantum problems
- 6:44Self-clocking with round-robin
- 8:07How SCRR works
- 10:38Bursty flow optimizations
- 12:25SCRR vs DRR performance
- 13:28Scheduling paradigms comparison
- 14:33Fairness and deployment potential
- 15:37Summary & Q&A
related talks
25mApproximating Fair Queueing on Reconfigurable Switches
16mFRCC: Towards Provably Fair and Robust Congestion Control
22mProgrammable Calendar Queues for High-speed Packet Scheduling
16mTowards Optimal Rack-scale μs-level CPU Scheduling through In-Network Workload...
18mCCC: Re-architecting Delay-based Congestion Control in Datacenter Networks
25mResQ: Enabling SLOs in Network Function Virtualization
13mSifter: An Inversion-Free and Large-Capacity Programmable Packet Scheduler
19mSwCC: Software-Programmable and Per-Packet Congestion Control in RDMA Engine