Doctoral Defense

Towards Low-Latency Computing Services in Heterogeneous Edge Environments

Yu Liu

August 8, 2024
9:30 AM
Light Engineering, Room 250
Advisor:  Yuanyuan Yang

The development of information technologies has facilitated many novel real-time applications, such as autopilot, smart home systems, and virtual reality gaming. These applications require real-time processing and handling of high-volume data. Cloud computing, which involves transmitting high-volume data between end users and the cloud, can result in high latency. Edge computing, which pushes computing power closer to end users, is a promising solution to support such applications. However, providing low-latency computing services in edge environments presents new challenges due to the inherent nature of edge networks, such as resource/energy limitations and device/network heterogeneity.

To minimize service latency in heterogeneous edge environments, this dissertation presents a comprehensive service placement and resource management mechanism. The proposed method is a game-theoretic-based approach with a provable approximation ratio and polynomial time complexity. Additionally, there is a tradeoff between the approximation ratio and the time complexity.

Building on the game-theoretic-based approach, the dissertation introduces a deep-learning-assisted service placement and resource management approach to enable fast decision-making for critical applications requiring low decision-making time, like telemedicine and autopilot. This approach leverages a deep neural network to
approximate the game-theoretic-based method.

Furthermore, this dissertation explores energy-aware edge computing to balance overall service latency and energy consumption. A stochastic approach is developed to minimize service latency while adhering to a time-average energy cost budget, offering a provable performance guarantee under non-stationary system states.