Abstract

The IoT revolution is driven by a surge of applications based on intelligent edge devices, such as smart cities and homes, autonomous vehicles, online video gaming, virtual and augmented reality, and machine learning. The edge devices generate large amounts of data and are the end-point executing user-facing applications. The back-end of these systems is a central service where data is continuously collected, aggregated, and analyzed, and a significant part of the logic is executed. Unfortunately, traditional cloud services cannot provide performance guarantees that these latency-critical applications require, mainly due to round-trip times which can take several hundreds of milliseconds. Thus, such applications will likely rely on Edge services–an emerging technology tightly coupled with 5G cellular networks. Edge solutions are envisioned to provide storage and compute infrastructure at interoperating edge nodes located at geographical locations near the users, such as the base of cellular network towers.

This project focuses on enabling unstable edge nodes to collectively provide a reliable storage service for unpredictable user demands. Towards this goal, redundant storage schemes for edge systems will be designed and evaluated. The schemes will then be implemented and validated in a system-level platform that combines a physical test-bed and large-scale simulations. The project’s primary goal is to characterize the schemes’ system-level costs and benefits and leverage this understanding to design an edge service that will be readily applicable to real settings. This research amalgamates system-level design and optimization with mathematical analysis and thus will stimulate students’ interest in both practice and theory associated with modern computing environments. In addition, this research will help improve the efficiency and robustness of edge computing, thus avoiding routing traffic to data centers. Such Edge services will reduce energy consumption, alleviate some of the load on the Internet backbone, and thus accelerate the development and adaptation of critical IoT applications.