Tuesday, April 6, 2021 -
The network stack in today's operating systems is a remnant from a time when a server had a handful of cores and processed requests from a few thousand clients. It simply cannot keep up with the scale of modern servers and the requirements of modern applications. Specifically, real-time applications and high user expectations enforce strict performance requirements on the infrastructure. Further, there is a fundamental shift in the way hardware capacity scales from simply relying on Moore's law to deliver faster hardware every couple of years to leveraging parallel processing and task-specific accelerators. This talk covers innovations in three key components of the network stack. First, I will cover my work on scalable packet scheduling in software network stacks, improving the control of traffic outgoing from large-scale servers. Second, I will move on to my work on improving overload control for servers handling microsecond-scale remote procedure calls, providing better control over incoming traffic to large-scale servers. Then, the talk covers my work on Wide Area Network (WAN) congestion control, focusing on network-assisted congestion control schemes, where end-to-end solutions fail. The talk will conclude with a discussion of plans for future research in this area.
Ahmed is a postdoctoral associate at MIT working with Prof. Mohammad Alizadeh. His research interests broadly cover the theory, design, and implementation of scalable computer networks and systems, including resource scheduling, congestion control, wireless networks, and cyber-physical systems. Before joining MIT, Ahmed received his PhD in computer science from Georgia Tech, where he was advised by Prof. Mostafa Ammar and Prof. Ellen Zegura. His PhD was partially supported by the Google PhD Fellowship in Systems and Networking. He received his bachelor's degree from Alexandria University in 2010.