Twenty Years After: Hierarchical Core-Stateless Fair Queueing
summary
Implements Core-Stateless Fair Queueing on programmable switches and generalizes it to Hierarchical CSFQ that approximates hierarchical fair queuing while keeping only interior-node state. Design uses a window-based rate estimator, discretized probabilistic dropping and a few recirculation packets to update fair share; it runs at line rate using a single FIFO queue. Testbed with 40 Gbps servers shows tenant isolation (about 3x per-flow throughput in the demo) and up to 60% lower flow completion time vs. TCP.
AI-written summary · about
chapters (10)
- 0:00Intro
- 1:12Core-Stateless Fair Queueing (CSFQ)
- 3:51Hierarchical Fair Queueing
- 4:28Hierarchical CSFQ
- 5:04Hierarchical Computation
- 7:00Data Plane Design Rate estimator
- 8:48Evaluation
- 9:31Testbed Experiments
- 10:23Large-scale DCN simulations
- 11:20Conclusion
related talks
25mApproximating Fair Queueing on Reconfigurable Switches
16mManaging Congestion Control Heterogeneity on the Internet with Approximate Performance...
16mGearbox: A Hierarchical Packet Scheduler for Approximate Weighted Fair Queuing
22mProgrammable Calendar Queues for High-speed Packet Scheduling
20mHarmony: A Congestion-free Datacenter Architecture
16mBuilding A CSFQ-Inspired Transport for Switched CXL Memory Pooling
22mEiffel: Efficient and Flexible Software Packet Scheduling
16mFlow Scheduling with Imprecise Knowledge
