Resource management for cost-effective cloud and edge systems

Abstract

With the booming of Internet-based and cloud/edge computing applications and services,datacenters hosting these services have become ubiquitous in every sector of our economy which leads to tremendous research opportunities. Specifically, in cloud computing, all data are gathered and processed in centralized cloud datacenters whereas in edge computing, the frontier of data and services is pushed away from the centralized cloud to the edge of the network. By fusing edge computing with cloud computing, the Internet companies and end users can benefit from their respective merits, abundant computation and storage resources from cloud computing, and the data-gathering potential of edge computing. However, resource management in cloud and edge systems is complicated and challenging due to the large scale of cloud datacenters, diverse interconnected resource types, unpredictable generated workloads, and a range of performance objectives. It necessitates the systematic modeling of cloud and edge systems to achieve desired performance objectives.

This dissertation presents a holistic system modeling and novel solution methodology to effectivelysolve the optimization problems formulated in three cloud and edge architectures: 1) cloud computing in colocation datacenters; 2) cloud computing in geographically distributed datacenters; 3) UAV-enabled mobile edge computing. First, we study resource management with the goal of overall cost minimization in the context of cloud computing systems. A cooperative game is formulated to model the scenario where a multi-tenant colocation datacenter collectively procures electricity in the wholesale electricity market. Then, a two-stage stochastic programming is formulated to model the scenario where geographically distributed datacenters dispatch workload and procure electricity in the multi-timescale electricity markets. Last, we extend our focus on joint task offloading and resource management with the goal of overall cost minimization in the context of edge computing systems, where edge nodes with computing capabilities are deployed in proximity to end users. A nonconvex optimization problem is formulated in the UAV-enabled mobile edge computing system with the goal of minimizing both energy consumption for computation and task offloading and system response delay. Furthermore, a novel hybrid algorithm that unifies differential evolution and successive convex approximation is proposed to efficiently solve the problem with improved performance.

This dissertation addresses several fundamental issues related to resource management incloud and edge computing systems that will further in-depth investigations to improve costeffective performance. The advanced modeling and efficient algorithms developed in this research enable the system operator to make optimal and strategic decisions in resource allocation and task offloading for cost savings.