Loading Events
This event has passed.

» Join the defence online (Zoom)


M.Sc. Pengyuan Zhou defends his doctoral thesis “Edge-Fa­cil­it­ated Mo­bile Com­put­ing and Com­mu­nic­a­tion” on Thursday the 28th of May 2020 at 12 o’clock noon. His opponent is Professor Tarik Taleb (Aalto University) and custos Professor Jussi Kangasharju (University of Helsinki). The defence will be held in English.



The proliferation of IoT devices and rapidly developing wireless techniques boost the data volume and service demand at the edge of the Internet. Meanwhile, increased requirement for low latency feedback has become a must for most popular mobile applications, e.g., Augmented Reality (AR), Virtual Reality (VR) and Connected Vehicles. To address these challenges, edge computing has emerged as an extensional solution for cloud computing.

This thesis studies edge computing-facilitated mobile computing and communication systems. We first propose solutions to improve edge resource utilization regarding general edge systems. We present a mechanism to cluster user requests based on similarity for better Content Delivery Network (CDN) performance. This mechanism works directly on current CDN architecture and can be deployed incrementally. Then we extend the mechanism by adding cache resource grouping algorithm, so that the system directs similar requests to same servers and group those servers which receive similar requests. This iterative mechanism optimizes the edge utilization by concentrating the resource on similar requests to achieve higher cache hit ratio and computation efficiency.

Thereafter, we present solutions for mobile edge systems specifically for three most promising use cases, i.e., Connected Vehicles, Mobile AR (MAR) and Smart city (traffic control). We explore the potential of edge computing in connected vehicular AR applications with real data sets. We design a lightweight edge system and data flow fit for general connected vehicular AR applications and implement a prototype. With an indoor test and real data set analysis, we find out that our system can improve the performance of vehicular AR applications with reasonable cost. To optimize the system, we formulate the problem of edge server allocation and task scheduling as a mutant multiprocessor scheduling problem and develop a two-stage edge-cloud decentralized algorithm as well as a centralized algorithm to schedule the offloading tasks on the fly. We conduct a raw road test and an extensive evaluation based on the road test results and large data sets from real world. The results show that our system improve at least twice the application performance comparing with cloud solutions.

For MAR, we consider to offload tasks to multiple edge servers via multiple paths simultaneously to further improve the MAR performance. We develop a fast scheduling algorithm to split the workloads among the available edge servers and show promising results with real implementations. At last, we explore the potential of combining edge computing and machine learning techniques to realize intelligent traffic control by letting edge servers co-located with traffic lights learn the waiting traffic and adapt the light periods with reinforcement learning.