Location

KACB 2100

Title: Systems Techniques for Enhancing Content Delivery in the Face of Network Variability in the Edge-Cloud Continuum

Date: 31st July 2024 (Wednesday)
Time: 10 AM - 12 PM ET
Location: KACB 2100
Virtual Meeting: Teams

 

Manasvini Sethuraman
Ph.D. Student
School of Computer Science
Georgia Institute of Technology

Committee:
Dr. Umakishore Ramachandran (Advisor) - School of Computer Science, Georgia Institute of Technology
Dr. Ashutosh Dhekne - School of Computer Science, Georgia Institute of Technology
Dr. Ragupathy Sivakumar - Electrical and Computer Engineering, Georgia Institute of Technology
Dr. Anand Sivasubramaniam - Dept. of Computer Science & Engineering, Pennsylvania State University
Dr. Ellen Zegura - School of Computer Science, Georgia Institute of Technology

Abstract:
Today's internet-based applications involve end users, located at the edge of the network, communicating with services, often hosted in cloud data centers. Connecting the two ends is the wide area network (WAN). The last hop on the WAN, to the end user is termed the last mile of internet connectivity and spans a variety of technologies ranging from wired high-speed broadband to cellular networks, wireless ISPs, and community mesh networks. Of particular interest are the wireless means of last-mile connectivity, that exhibit temporal and spatial variability in the available capacity (bandwidth). This variable capacity is often the root cause of poor/variable user experience with applications like video conferencing, on-demand video streaming, and other bandwidth-intensive and/or latency-sensitive applications. Many application-specific solutions exist today, mainly for use with on-demand video streaming in wireless last-mile cellular links, that may not generalize well across applications. Applications typically consist of client and server components, residing on user devices and cloud data centers respectively. Many solutions for deploying the server end of these applications have been developed with data center wired networks in mind. However, when we consider the space of community mesh networks, which, in addition to providing internet connectivity, often host local services such as file servers,  and video conferencing, existing techniques for application scheduling and orchestration for the server-side components may not be a good fit since they often do not consider the effect of bandwidth variability.

In this dissertation, we elevate available bandwidth as a first-class citizen in resource orchestration and design a holistic approach to the management of available bandwidth. We present three techniques for addressing bandwidth variations at different levels in the system software encompassing the end devices and the computing infrastructure:

  • At the level of a mobile device, we demonstrate ways to manage the available bandwidth among multiple user applications by creating a system-wide bandwidth management service to handle spatiotemporal variations in bandwidth under conditions of user mobility.
  • From the perspective of an application service provider (e.g., Netflix/ YouTube), we consider the use of emerging technologies to effectively deliver content at the edge to mobile users, and at the same time, reduce reliance on cellular last mile data, using a combination of content prefetching and high speed out-of-band mmWave links.
  • At the level of shared network infrastructure, comprising multiple applications, we address the problem of scheduling applications in an environment where the communication links are predominantly wireless. We show the importance of bandwidth-aware scheduling and migration of application components towards the application's performance and user experience.