Semi-Oblivious Traffic Engineering: The Road Not Taken
summary
Combines Räcke’s oblivious routing for a static, capacity-aware path set with a centralized LP that dynamically adjusts split ratios; a software framework models and simulates this semi‑oblivious TE on Facebook’s backbone and many ISP topologies. Results show near‑optimal throughput and robustness, delivering about 100% of traffic with maximum congestion within 16% of optimal using a path budget of four (others need ~32 paths to match).
AI-written summary · about
chapters (7)
- 0:00Traffic engineering overview
- 2:12Limitations of existing models
- 5:21The case for static paths
- 7:29Path selection strategies
- 13:01Introduction to SMALL
- 16:07Evaluation and results
- 18:53Conclusion and Q&A
related talks
17mDOTE: Rethinking (Predictive) WAN Traffic Engineering
17mPrecise Data Center Traffic Engineering with Constrained Hardware Resources
17mTraffic Engineering for AI Training Networks
25mVirtualizing Traffic Shapers for Practical Resource Allocation
16mExpress Lane to Efficiency and Reliability: Multi-Dimensional Control in Meta’s Express..
13mAUTO: Adaptive Congestion Control Based on Multi-Objective Reinforcement Learning...
13mEfficient Navigation for Modern Networks
17mTowards provably performant congestion control
