DOTE: Rethinking (Predictive) WAN Traffic Engineering

on YouTube ↗17m

" summary

Claims a system called DOTE that trains a deep neural network to map recent demand history to a WAN traffic engineering configuration, avoiding demand prediction and LP solving, and outputs splitting ratios and rate caps. Uses stochastic normalized gradient ascent and proves convergence in standard models. Empirical results on Microsoft production traces show TE quality nearly matches an omniscient oracle and beats prediction baselines (demand prediction is ~3× worse at the 99th percentile, DOTE is 32% over optimum). Code is on GitHub and runtime is substantially lower.

AI-written summary · about

" chapters (6)

1. 0:00Start
2. 2:24A Tale of Two Traffic Patterns
3. 5:52Intuition for DOTE's offline training
4. 9:52Extending DOTE to other optimization objectives
5. 13:00Minimizing Maximum Link Utilization
6. 14:44Performance under link failures (for PWAN)

" related talks

• 17mTraffic Engineering for AI Training NetworksShuqiang Zhang, Jingyi Yang · Networking @Scale 2023
• 17mPrecise Data Center Traffic Engineering with Constrained Hardware ResourcesRui Wang, Srinivasan Seshan, Peter Steenkiste, Keqiang He, Shawn Shuoshuo Chen · 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI '24)
• 13mA Fast Solver-Free Algorithm for Traffic Engineering in Large-Scale Data Center NetworkYingming Mao, Qiaozhu Zhai, Ximeng Liu, Zhen Yao, Xia Zhu, Yuzhou Zhou · NSDI '26
• 21mSemi-Oblivious Traffic Engineering: The Road Not TakenPraveen Kumar, Petr Lapukhov, Robert Soulé, Nate Foster, Robert Kleinberg, Chiun Lin Lim, Yang Yuan, Chris Yu · NSDI '18
• 17mReasoning about Network Traffic Load Property at Production ScaleHao Wu, Dennis Cai, Ennan Zhai, Chenren Xu, Ruihan Li, Fangdan Ye, Yifei Yuan, Ruizhen Yang, Bingchuan Tian, Tianchen Guo, Xiaobo Zhu, Zhongyu Guan, Qing Ma, Xianlong Zeng · 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI '24)
• 13mEfficient Multi-WAN Transport for 5G with OTTERSharad Agarwal, Ryan Beckett, Paramvir Bahl, Rachee Singh, Mary Hogan, Gerry Wan, Yiming Qiu · NSDI '25
• 12mATP: In-network Aggregation for Multi-tenant LearningMichael Swift, Aditya Akella, Wenfei Wu, ChonLam Lao, Kshiteej Mahajan, Yanfang Le, Yixi Chen · NSDI ’21
• 14mBrain-on-Switch: Towards Advanced Intelligent Network Data Plane via NN-Driven Traffic...Qi Li, Jinzhu Yan, Haotian Xu, Zhuotao Liu, Ke Xu, Mingwei Xu, Jianping Wu · 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI '24)