Date of Award

Spring 5-31-2018

Document Type


Degree Name

Doctor of Philosophy in Electrical Engineering - (Ph.D.)


Electrical and Computer Engineering

First Advisor

Nirwan Ansari

Second Advisor

Ashutosh Dutta

Third Advisor

Edwin Hou

Fourth Advisor

Abdallah Khreishah

Fifth Advisor

MengChu Zhou


Internet scale data centers are generally dispersed in different geographical regions. While the main goal of deploying the geographically dispersed data centers is to provide redundancy, scalability and high availability, the geographic dispersity provides another opportunity for efficient employment of global resources, e.g., utilizing price-diversity in electricity markets or utilizing locational diversity in renewable power generation. In other words, an efficient approach for geographical load balancing (GLB) across geo-dispersed data centers not only can maximize the utilization of green energy but also can minimize the cost of electricity. However, due to the different costs and disparate environmental impacts of the renewable energy and brown energy, such a GLB approach should tap on the merits of the separation of green energy utilization maximization and brown energy cost minimization problems. To this end, the notion of green workload and green service rate, versus brown workload and brown service rate, respectively, to facilitate the separation of green energy utilization maximization and brown energy cost minimization problems is proposed. In particular, a new optimization framework to maximize the profit of running geographically dispersed data centers based on the accuracy of the G/D/1 queueing model, and taking into consideration of multiple classes of service with individual service level agreement deadline for each type of service is developed. A new information flow graph based model for geo-dispersed data centers is also developed, and based on the developed model, the achievable tradeoff between total and brown power consumption is characterized.

Recently, the paradigm of edge computing has been introduced to push the computing resources away from the data centers to the edge of the network, thereby reducing the communication bandwidth requirement between the sources of data and the data centers. However, it is still desirable to investigate how and where at the edge of the network the computation resources should be provisioned. To this end, a hierarchical Mobile Edge Computing (MEC) architecture in accordance with the principles of LTE Advanced backhaul network is proposed and an auction-based profit maximization approach which effectively facilitates the resource allocation to the subscribers of the MEC network is designed. A hierarchical capacity provisioning framework for MEC that optimally budgets computing capacities at different hierarchical edge computing levels is also designed. The proposed scheme can efficiently handle the peak loads at the access point locations while coping with the resource poverty at the edge. Moreover, the code partitioning problem is extended to a scheduling problem over time and the hierarchical mobile edge network, and accordingly, a new technique that leads to the optimal code partitioning in a reasonable time even for large-sized call trees is proposed. Finally, a novel NOMA augmented edge computing model that captures the gains of uplink NOMA in MEC users' energy consumption is proposed.