Energy Efficiency And Load Balancing In Mobile Edge Computing

The phenomenal growth of the emerging computationally intensive ap-plications(such as image processing,automated driving vehicles,augmented reality,and automated driving)is posing unprecedented challenges on the pro-cessing capability,latency and reliability of future networks.By enabling cloud-computing capability at the network edge close to mobile devices,mo-bile edge computing(MEC)can provide pervasive computing services to meet the requirements of low-latency,high reliability,and high connectivity of the computationally demanding applications.MEC is also one of the key tech-nologies of the fifth generation mobile communication(5G).Energy efficiency and load balancing are key research topics in MEC and critical for the data processing capability and user experience of MEC.This thesis focuses on the researches of energy efficiency and load bal-ancing in MEC.For energy efficiency of MEC,both single-cell and multi-cell scenarios are considered,and online learning resource management approaches in dynamic environments are proposed to maximize the energy efficiency for energy-constrained mobile devices and networks,respectively.For load bal-ancing of MEC,the ubiquitous computing framework is considered,and online distributed approaches to balance the loads in computing and storage between edge and cloud servers are proposed.The proposed approaches can optimize task processing decisions based on task characteristics and network load adap-tively to achieve the advantages of both edge computing and cloud computing.The novelties and contributions of this thesis are summarized as follows.For energy efficiency in single-cell MEC,the BS can only have outdated knowledge of device channel conditions due to the time-varying nature of prac-tical wireless channels.To this end,a hybrid learning approach is proposed to optimize the real-time local processing and predictive computation offloading decisions in a distributed manner.Stochastic gradient descent(SGD)and on-line convex optimization(OCO)are integrated in the primal-dual optimization framework.The proposed hybrid learning approach can be decentralized be-tween the BS and mobile devices for scalability to maximize the time-average energy efficiency of mobile devices.We prove the asymptotic optimality of the proposed hybrid learning approach,where the optimality loss resulting from the differently-aged network states can diminish with the decreasing stepsizes of SGD and OCO.For energy efficiency in multi-cell MEC,an energy-efficient resource man-agement approach is developed based on distributed online learning to tackle the heterogeneity of computing and wireless transmission capabilities of edge servers and mobile devices.The proposed approach optimizes the decisions on task offloading,processing,and result delivery between edge servers and mo-bile devices to maximize the time-average energy efficiency of MEC.SGD is exploited to decouple the optimizations between time slots.A graph matching problem is then formulated for every time slot by decoupling and unifying the non-uniform cardinalities,and solved in a distributed manner by developing a new linear time complexity method.We prove that the optimality loss resulting from the distributed approximate graph matching method can be compensated and diminish by increasing the learning time.For load balancing of computing resources under ubiquitous computing framework,a distributed online collaborative computing approach is proposed based on Lyapunov optimization for data analysis in IoT application to minimize the time-average energy consumption of network.The arriving IoT data can be offloaded adaptively to edge servers to be pre-processed and the pre-processed results need to be transmitted to the cloud data center for analysis.Given the finite buffer of IoT devices,the biases(namely,"virtual placeholders")of the buffers are designed to create sufficient backlogs based on the network topology,link capacities,and server processing capabilities.The optimization of the placeholders is proved to be a new three-layer shortest path problem,and solved in a distributed manner by extending the Bellman-Ford algorithm.The proposed approach can significantly increase network throughput and reduce end-to-end delays.Lastly,for load balancing of storage under ubiquitous computing frame-work,a distributed content delivery approach based on online learning is pro-posed for caching application.A new profitable cooperative region is estab-lished for every content request admitted at an edge server,to avoid the invalid request dispatching.SGD is exploited to decouple the problem over time.The cooperative region for a content is erected at individual servers by comparing the upper and lower bounds for the backlogs of unsatisfied requests of the con-tent.The approach can operate in conjunction with simple caching policies of individual servers,to improve the efficiency of content placement,content delivery,and quality of experience for users.