HEDGE: Traffic Engineering with Probabilistic Link Capacities
summary
Lab experiments and a production cloud trace show that individual wavelengths on the same fiber fail independently enough to make link capacity a discrete random variable, causing up to 40 capacity fluctuations per five minutes. The authors build a 100 km optical testbed with three wavelengths and quantify failures across QPSK/8QAM/16QAM. They introduce wavelength-specific faults, implement HEDGE-A provisioning to meet Cmin with the fewest spectral channels, and run HEDGE TE to maximize expected throughput, dropping 12.2× fewer flows and displacing tunnels 622× less than an oracle reconfigure approach.
AI-written summary · about
chapters (7)
- 0:00HEDGE introduction
- 2:05Wavelength faults in optical links
- 4:55Stochastic link capacities evidence
- 5:58HEDGE capacity guarantees
- 9:50HEDGE TE for stochastic networks
- 13:24HEDGE performance evaluation
- 15:31Broader implications
related talks
31mExpectAll: A BDD Based Approach for Link Failure Resilience in Elastic Optical Networks
32mLive Q+A 2
23mACK to the Future: TCP in 2025
16mOptimizing Network Provisioning through Cooperation
18mLive Group Q&A Session
17mReasoning about Network Traffic Load Property at Production Scale
25mVirtualizing Traffic Shapers for Practical Resource Allocation
26mF10: A Fault-Tolerant Engineered Network
