Date of Award

12-2019

Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Duc A. Tran

Second Advisor

Dan Simovici

Third Advisor

Bo Sheng

Abstract

Mobile Edge Computing (MEC) has emerged as a viable technology for mobile operators to push computing resources closer to the users so that requests can be served locally without long-haul crossing of the network core, thus improving network efficiency and user experience. In MEC, commodity servers are deployed in the edge to form a distributed network of mini datacenters. A consequential task is to partition the user cells into groups, each to be served by an edge server, to maximize the offloading to the edge. The conventional setting for this problem in the literature is: (1) assume that the interaction workload between two cells has a known interaction rate, (2) compute a partition optimized for these rates, for example, by solving a weighted-graph partitioning problem, and (3) for a time-varying workload, incrementally re-compute the partition when the interaction rates change. This setting is suitable only for infrequently changing workloads. The operational and computation costs of the partition update can be expensive and it is difficult to estimate interaction rates if they are not stable for a long period. Hence, this dissertation is motivated by the following questions: is there an efficient way to compute just one partition, no update needed, that is robust for a highly time-varying workload? Especially, what if we do not know the interaction rates at any time? By ``robust", we mean that the cost to process the workload at any given time remains small despite unpredictable workload increases. Another consideration is geographical awareness. The edge servers should be geographically close to their respective user cells for maximizing the benefits of MEC. This dissertation presents novel solutions to address these issues. The theoretical findings are substantiated by evaluation studies using real-world data.

Share

COinS