Stochastic optimizations of mobile energy management

With the growing computing power of mobile devices and increasingly sophisticated applications, mobile computing and wireless communication are nowadays pervasive. However, since mobile devices have limited energy storage and sporadic energy supply, energy management remains a critical issue in mobile networks. To address the problem, a few approaches have been introduced to develop intelligent and optimal energy management for mobile networks. Firstly, wireless energy charging can be employed to replenish batteries of mobile devices. The mobile devices can harvest or receive energy to charge its battery without being physically connected to any power source. Alternatively, to control the energy consumption and resource usage, a mobile device can offload energy-intensive jobs to other devices, e.g., cloudlets. This thesis aims to address some of the important issues of energy management in mobile networks with wireless energy charging and job offloading. Three major contributions are presented in this thesis. Firstly, with the wireless energy transfer and harvesting technologies (e.g., radio frequency, or namely RF energy), mobile devices are fully untethered as energy supply is more ubiquitous. The mobile devices can receive energy from wireless chargers which can be static or mobile. We introduce the use of a mobile energy gateway that can receive energy from a fixed charging facility, move, and transfer the energy to other mobile users. The mobile energy gateway aims to maximize the utility by taking energy charging/transferring actions optimally. We formulate an optimal energy charging/transferring problem as a Markov decision process (MDP). The MDP model is then solved to obtain an optimal energy management policy for the mobile energy gateway. Furthermore, we prove that the optimal energy management policy has a threshold structure. We conduct an extensive performance evaluation of the MDP-based energy management scheme. The proposed MDP-based scheme outperforms several conventional baseline schemes in terms of expected overall utility. Secondly, we develop an optimal energy charging scheme for the mobile device, considering the states of location, energy storage, as well as stochastic traffic generation which determines energy demand. In this case, energy management takes data flows in the form of traffics (i.e., job processing and data transmission) into consideration. We formulate the energy charging problem as an MDP to obtain the mobile device's optimal policy. The objective is to maximize the expected utility. Additionally, we prove that the optimal policy of the proposed MDP has a threshold structure. The numerical results show the optimality of the proposed MDP-based wireless energy charging scheme compared with baseline schemes under various scenarios and parameter setting. With a different approach, the emergence of mobile cloud computing enables mobile devices to offload applications to nearby mobile resource-rich devices (i.e., cloudlets) to reduce energy consumption and improve performance. However, because of mobility and limited cloudlet capacity, the connections between a mobile device and cloudlets can be intermittent. Thus, offloading actions taken by the mobile device may fail. We therefore develop an optimal offloading algorithm for the mobile device in the intermittently connected cloudlet system, considering the local load on the mobile device and availability of cloudlets. We examine mobility patterns of the mobile device and admission control of cloudlets, and then derive the probability of successful offloading actions analytically. We formulate and solve an MDP model to obtain an optimal policy for the mobile device with the objective to minimize the computation and offloading costs in terms of energy and resource consumption. We prove that the optimal policy has a threshold structure. Subsequently, we also introduce a fast algorithm for an energy-constrained mobile device to make offloading decisions. In summary, this thesis investigates a few important energy management problems in mobile networks. Wireless energy charging is employed to replenish the battery storage of mobile devices. Moreover, mobile cloud computing is used for the mobile device to offload jobs to control the energy and resource consumptions. MDP-based schemes are employed as efficient energy managing approaches to obtain the optimal policies in the networks. The obtained optimal policies are shown to outperform baseline schemes.DOCTOR OF PHILOSOPHY (SCE