Essays on User-Preference-Based Optimization Methods for Overcoming the Resource Limitations of Smartphones

Abstract

There is no doubt that the realized technological development in smartphones technology and its applications during the last few years had resulted in a remarkable increase in the smartphone usage, not only for simple services such as calling and sending messages, but also for performing a huge number of activities such as social networking, web browsing, emailing, video watching, and gaming. However, running such activities implies a heavy workload on smartphones and results on significant resources consumption (memory, processing, communication, and energy). Despite its attractive design (i.e. small size, and light weight), diverse range of capabilities, and multi-functionality, smartphone still have limited resources, such as battery energy, processing power, network bandwidth, and storage capacity. These constraints of mobile resources result in limited use of smartphones and boost the need for solutions to overcome these limitations. Battery lifetime of smartphone is one of the most critical limitations of smartphones. It has always been a concern not only for smartphone users but also for the manufacturers. Despite the significant efforts to improve battery technology, the advance in the battery technology has not been able to keep pace with the rapidly growing demand for power consumption for mobile activities. Because smartphone battery lifetime depends on usage behavior, it is necessary to understand how users use up the battery of the smartphone. Many researchers have made great efforts on usage behavior of smartphone users. All of them found that users have their own usage pattern. This diversifies among users revealed the need of understanding the user behavior. Without understanding user behavior, it is not possible to clear understand the impact of any optimization on user experience or the mobile power consumption. Recently, Mobile cloud computing (MCC) has been introduced to overcome the resource restrictions of smartphones by enabling them to offload the computational intensive applications on resourceful clouds instead of running them locally on the device. This is referred to as computation offloading. However, offloading the computation to the cloud definitely introduces costs for the mobile user, based on the exact resources consumed (energy

bandwidth

money

etc.). Therefore, offloading should only occur if the offloading cost is smaller than the cost of local. In addition to the cost issues, there are some user context issues such as the remaining energy in his smartphone, and his location, which does not make it always desirable to offload the application execution to the cloud. The overall objective of this study is to find solutions for overcoming the limitations of resource-constraint smartphones considering two perspectives: energy consumption, and smartphone applications execution. To address these resources restrictions of smartphone, this dissertation focuses on two key issues. First: the energy consumption optimization was studied by taking into account the user preferences to maximize the users utility from the energy remaining in his smartphone battery. In particular, a utility-based energy consumption optimization model was proposed which considers the user preferences with respect to energy allocation to the two types of smartphone application uses (i.e., on-device application use and cloud-based application use). The working of the model is demonstrated by conducting both quantitative and qualitative research techniques including collecting usage data from real smartphone users, and in-person interview. Second issue deals with multi-criteria optimization for supporting computation offloading of smartphone applications to the mobile cloud computing (MCC). Here, an offloading decision making model that minimizes the offloading cost subject to multiple constraints including application, smartphone, network, and cloud constraints while considering the user context and the user preferences is proposed. The different costs (i.e., time cost, energy cost, communication cost, and computation cost), which are incurred by executing the application locally on the mobile device or by offloading the application to the cloud), are integrated through the use of user preferences in making this multi-criteria offloading decision. Although it can be assumed that users can consider all various combinations of those factors and make a good decision, the frequency of those decisions will be cumbersome for the user. In addition, there is no-specific linear equation which can describe the relationship between all those factors that influence the offloading decision. In this regard, the neural networks are used to model the non-linear interaction among these multiple factors. A Deep Neural Network (DNN) was trained using offloading decisions examples made by the user based on the proposed offloading decision model to support the user decision in making optimal offloading decision in the long run. The potential of neural networks for making offloading decision is evaluated through a use case. The main contribution of this research rises from the new approaches that this work presents in dealing with the overcoming the resource restrictions of smartphone. Most of the previous studies about smartphone usage and energy consumption as well as the computation offloading are frequently focusing on developing technical solutions and rarely consider the user as a part of their solutions. Such solutions unable to capture user intention and preferences and hence fail to match the user expectations from using the smartphone. Therefore, this work opens the doors for new research area in smartphone technology that will contribute to further improvements in the smartphone platforms.