Solving Max-Min Fair Resource Allocations Quickly on Large Graphs

Explore a cutting-edge conference talk on solving max-min fair resource allocation problems efficiently for large-scale graphs. Dive into innovative approaches that improve upon existing solutions, including a method to convert optimization sequences into a single fast optimization and a generalization of waterfilling for multi-path scenarios. Discover how these new algorithms Pareto-dominate previous techniques, delivering faster, fairer, and more efficient allocations. Learn about the theoretical guarantees of some allocators and their practical implementation in Azure's WAN traffic engineering pipeline, resulting in significant speedups while maintaining solution quality. Gain insights into the potential impact of these advancements on WAN traffic engineering and cluster scheduling, particularly for larger problem sizes.

NSDI '24 - Solving Max-Min Fair Resource Allocations Quickly on Large Graphs