Optimization Research Of Performance And Energy Consumption For Mobile Network Application

Our mobile network has developed into the universal era.The mobile network application market continues to grow under the demand stimulation of the huge user group.On the one hand,the high-quality wireless communication environment and high-performance CPU provide a good platform for the development of mobile network applications at the same time,guarantee a good user experience.On the other hand,the increasingly traffic load and unreasonable resource scheduling strategy of mobile network applications will undoubtedly to increase the burden of mobile battery,which will hinder the development of mobile network products and service market.Because of the huge market share and potential of mobile network,the mobile network applications' design and optimization are the research focus in academic and industry.At present,existed optimization schemes for mobile network applications prefer to measure and optimize the energy consumption in a static way,which rarely consider the dynamic influence of the network interface status,user experience,user patterns,network environment,context information and other factors of the energy consumption of mobile network applications.This dissertation focuses on the energy and performance optimization for the mobile network application download strategy and web browsing of mobile web browser.To this end,we design the optimized download algorithm by dynamic optimization method(Lyapunov optimization framework and MDP model)and build the resource scheduling model for web browsing by supervised learning algorithm(Support Vector Machine),the main content of this dissertation are as follows:1)Propose the Low power and Link selection download algorithm based on Lyapunov optimization frameworkThis dissertation studies the theory of Lyapunov optimization framework.Combine the stability of network queue with energy characteristics of different transmission links,this dissertation proposes the Low Power and Link Selection Algorithm(LLA)which aims at one-node optimization based on the Lyapunov drift.LLA handles the download request dynamically and selects the low power link to download data based on the current network environment,the communication quality and the information of download requests.LLA optimizes the energy and performance of data download while keeping the stability of network download request queue of mobile network applications.Experiments are conducted to verify the effectiveness of LLA.2)Propose the Energy-aware Download Model based on Markov Decision Process This dissertation studies the MDP theory and analyzes user usage patterns.Combining the energy management of data download problem with MDP and building the Energy-aware Download Model(EDM)based on the MDP for different user usage patterns.The model considers different user usage patterns during the decision-making process,and bases on the maximum reward principle to schedule the resources dynamically.The goal is to minimize the energy consumption of data download while improve user experience.Finally,we evaluate two adaptive download strategies(EDM and LLA)with the Android default download policy.The result proves that adaptive download strategies perform better in energy and performance.In addition,EDM outperforms LLA for light and regular users.3)Propose the Web browsing optimized model based on the Support Vector Machine This dissqrtation studies the web load process and analyzes the influence of web infrastructure and styles information for loading webpage.Build the web browsing optimized model on heterogeneous multiprocessors platform.The model combines the supervised learning algorithm SVM with the web load analysis and scheduling the web rendering work on the appropriate cores with right frequency based on different user requirements(fastest load time,least energy consumption and smallest EDP value).Optimize the web browsing while improving user experience by different personalized optimized models.Experiments are conducted to validate the adaptive scheduling approaches(WS and SVM model)which perform better than the system default strategy Linux HMP.In addition,the proposed SVM models perform better than WS with different metrics(energy,load time and EDP)In addition,this dissertation also studies the web browsing under different network environments.Analyze the load time proportion of network delay and CPU processing.Applies the appropriate SVM models which are based on network quality.The experiment verifies the effectiveness of SVM model in different network environments.