Characterizing and Improving Large-Scale Networked Systems with Network Measurements

Date: 2020/08/11 - 2020/08/11

Academic Seminar: Characterizing and Improving Large-Scale Networked Systems with Network Measurements

Speaker: Yibo Pi, University of Michigan

Time: 9:00-10:00 a.m. August 11th, 2020 (Tuesday)

Location:via Zoom (Meeting ID: 67881955411 Password: 4412)


Large-scale networked systems like Internet and distributed machine learning systems could include thousands and even billions of clients. To understand and manage such systems, it is crucial to develop scalable measurement and management approaches, and design thorough experiments for a complete characterization of the system.

In the talk, I will introduce the first data-driven approach to global load balancing in content delivery networks, which directs billions of clients to their nearby servers.

This approach continuously analyzes the streaming traffic data of service providers and thus dynamically adapts to changing network conditions. Using a real-world dataset provided by a commercial company, I demonstrate that this approach is more scalable and accurate than existing ones. I will also present that the performance imbalance issues of Internet load-balanced paths, previously deemed insignificant, are now both prevalent and significant on the Internet. To show the wide impacts of these issues, I experiment on several real-world applications and propose potential solutions for each application.

I will conclude this talk by discussing an ongoing project on federated learning and my future directions on Internet, distributed machine learning systems and network security.


Yibo Pi is a final-year Ph.D. student in Computer Science and Engineering at the University of Michigan. His research interests focus on measuring, improving and securing large-scale networked systems, with an emphasis on the Internet and distributed deep learning systems. During his Ph.D., he has found several performance issues in the Internet load balancing and has demonstrated the wide impacts of these issues on real-world applications. Before coming to the University of Michigan, he completed both his B.S. and M.S. in Electrical and Computer Engineering in the UM-SJTU Joint Institute.