ORCID ID

0000-0003-1967-4681

Date Awarded

2019

Document Type

Dissertation

Degree Name

Doctor of Philosophy (Ph.D.)

Department

Computer Science

Advisor

Qun Li

Committee Member

Peter Kemper

Committee Member

Gang Zhou

Committee Member

Zhenming Liu

Committee Member

Gexin Yu

Abstract

Distributed computing has never stopped its advancement since the early years of computer systems. In recent years, edge computing has emerged as an extension of cloud computing. The main idea of edge computing is to provide hardware resources in proximity to the end devices, thereby offering low network latency and high network bandwidth. However, as an emerging distributed computing paradigm, edge computing currently lacks effective system support. To this end, this dissertation studies the ways of building system support for edge computing. We first study how to support the existing, non-edge-computing applications in edge computing environments. This research leads to the design of a platform called SMOC that supports executing mobile applications on edge servers. We consider mobile applications in this project because there are a great number of mobile applications in the market and we believe that mobile-edge computing will become an important edge computing paradigm in the future. SMOC supports executing ARM-based mobile applications on x86 edge servers by establishing a running environment identical to that of the mobile device at the edge. It also exploits hardware virtualization on the mobile device to protect user input. Next, we investigate how to facilitate the development of edge applications with system support. This study leads to the design of an edge computing framework called EdgeEngine, which consists of a middleware running on top of the edge computing infrastructure and a powerful, concise programming interface. Developers can implement edge applications with minimal programming effort through the programming interface, and the middleware automatically fulfills the routine tasks, such as data dispatching, task scheduling, lock management, etc., in a highly efficient way. Finally, we envision that consensus will be an important building block for many edge applications, because we consider the consensus problem to be the most important fundamental problem in distributed computing while edge computing is an emerging distributed computing paradigm. Therefore, we investigate how to support the edge applications that rely on consensus, helping them achieve good performance. This study leads to the design of a novel, Paxos-based consensus protocol called Nomad, which rapidly orders the messages received by the edge. Nomad can quickly adapt to the workload changes across the edge computing system, and it incorporates a backend cloud to resolve the conflicts in a timely manner. By doing so, Nomad reduces the user-perceived latency as much as possible, outperforming the existing consensus protocols.

DOI

http://dx.doi.org/10.21220/s2-wvkj-fj82

Rights

© The Author

Download

Share Feedback

Share

COinS