IoT Thrust Seminar | Innovating Next-Generation Networks with Optical Technologies
Optical technologies are driving the next-generation networks to meet the diverse requirements of emerging applications. In this talk, I will present three representative projects: UCMP (SIGCOMM 2024), OTSS-PON (JOCN 2019 & 2022), and EchelonFlow (HotNets 2022). UCMP introduces a novel cost metric with uniform cost in optical data center networks, unifying the impacts of topology disruption and hop count on latency and bandwidth efficiency. OTSS-PON is a cutting-edge network architecture for metro-access networks, offering transparent, low-latency connections for latency-sensitive applications. EchelonFlow provides an innovative network abstraction for AI clusters, capturing the vastly different computation patterns across training paradigms. Finally, I will emphasize the growing importance of AI clusters for AI services, paralleling the role of data center networks in cloud computing over the past 15 years. I will conclude by briefly discussing how to design efficient architectures for AI training, realizing rapid and robust failure recovery, and supporting model training across large-scale networks.
Jialong Li is a postdoctoral researcher at Max Planck Institute for Informatics (MPI-INF), Germany. Before joining MPI-INF, he received his B.E. and Ph.D. degrees in Electronic Engineering from Tsinghua University in 2016 and 2021, respectively. His research focuses on computer networks and data center networks, with his work published in top venues such as SIGCOMM, ToN, JOCN, and OFC. He has authored over 20 academic papers and served on program committees for conferences and as a reviewer for several journals. He has received the Tsinghua University Comprehensive Excellence Scholarship and the Tsinghua University Friends - Xiaomi Scholarship.