Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about advanced network failure mitigation techniques in this 16-minute conference talk from NSDI '25 that presents a novel performance-aware ranking system for cloud datacenter networks. Discover how researchers from the University of Southern California and Microsoft developed innovative methods to optimize end-to-end flow-level metrics rather than relying on traditional local criteria or global proxy metrics for network failure response. Explore the technical approach that enables quick estimation of mitigation impact and high-fidelity ranking of different response actions, supporting a broader range of mitigation strategies through holistic analysis. Examine real-world results from a large cloud provider demonstrating orders of magnitude improvements in flow completion time and throughput, while understanding how this approach scales effectively to large datacenter environments. Gain insights into the limitations of existing network mitigation systems and how direct optimization of flow-level performance metrics can significantly enhance network reliability and performance in cloud infrastructure.
