Flow Scheduling with Imprecise Knowledge in Data Center Networks

Explore a groundbreaking approach to data center network flow scheduling in this 16-minute conference talk from NSDI '24. Delve into QCLIMB, a novel solution that minimizes flow completion times using imprecise flow information. Learn how the researchers from Tianjin University, Hong Kong University of Science and Technology, University of Science and Technology of China, Dalian University of Technology, and New York University Shanghai leverage machine learning techniques to estimate flow bounds accurately. Discover the two key components of QCLIMB: a scheduling algorithm that prioritizes small flows from the start of transmission and an efficient out-of-order handling mechanism. Gain insights into how QCLIMB outperforms existing solutions like PIAS and approaches the performance of pFabric without requiring switch modifications. Understand the potential impact of this research on improving data center network efficiency and performance.