We explore a new design point for traffic engineering on wide-area networks (WANs): directly optimizing traffic flow on the WAN using only historical data about traffic demands. Doing so obviates the need to explicitly estimate, or predict, future demands. Our method, which utilizes stochastic optimization, provably converges to the global optimum in well-studied theoretical models. We employ deep learning to scale to large WANs and real-world traffic. Our extensive empirical evaluation on real-world traffic and network topologies establishes that our approach's TE quality almost matches that of an (infeasible) omniscient oracle, outperforming previously proposed approaches, and also substantially lowers runtimes.
|Original language||American English|
|Title of host publication||Proceedings of the 20th USENIX Symposium on Networked Systems Design and Implementation, NSDI 2023|
|Number of pages||25|
|State||Published - 2023|
|Event||20th USENIX Symposium on Networked Systems Design and Implementation, NSDI 2023 - Boston, United States|
Duration: 17 Apr 2023 → 19 Apr 2023
|Name||Proceedings of the 20th USENIX Symposium on Networked Systems Design and Implementation, NSDI 2023|
|Conference||20th USENIX Symposium on Networked Systems Design and Implementation, NSDI 2023|
|Period||17/04/23 → 19/04/23|
Bibliographical noteFunding Information:
Acknowledgements: We thank our shepherd, Mojgan Ghasemi, and the NSDI reviewers, for their valuable feedback. We thank Umesh Krishnaswamy, Himanshu Raj and the SWAN team at Microsoft for their help and feedback. Yarin Perry and Michael Schapira were partially supported by BSF grant 2019798 and a grant from Microsoft. Aviv Tamar is funded by ERC grant 101041250.
© NSDI 2023.All rights reserved