Book Description

Cloud networks, consisting of multiple data centers that are distributed at different geographical locations and interconnected by wide-area networks, are emerging as the next generation platform of cloud computing, due to their great potential in providing elastic, flexible, and pervasive cloud services. The operation of such a cloud network usually incurs high operational cost, e.g., electricity to power its data centers. In this thesis we study resource allocation and provisioning in a cloud network to minimize operational costs or network delays, while meeting various user Service Level Agreements (SLAs). This however poses great challenges, since the cloud network provides various resources to meet the dyanmic resource demands of various users. The distributed cloud resources further complicate the resource allocations. Existing studies considered only resource allocations in a single data center or with a single SLA, and therefore are not directly applicable to the cloud network. The development of new techniques for cost-aware resource allocation and provisioning for cloud networks is desperately needed. This thesis will tackle these key issues as follows. We firstly deal with fair dispatching of user requests to different data centers in the cloud network by taking time-varying electricity prices and workloads of data centers into consideration, such that the operational cost of the cloud network is minimized. We propose an adaptive optimization framework, and devise a fast and fair approximation algorithm with a provable approximation ratio. We secondly study the resource allocation and provisioning in a cloud network, by exploring heterogeneities of cloud resources and user demands. For this problem, we propose a two-stage optimization framework: dispatching user task requests to different data centers; followed consolidating Virtual Machines in the same data center into different servers, and we devise efficient algorithms based on the proposed framework, such that the electricity cost of the cloud network is minimized, while meeting various aspects of user SLAs. We thirdly address in the cloud network by investigating the periodic resource demands of virtual networks. We propose an efficient embedding algorithm by incorporating a novel embedding metric that accurately models dynamic workloads among data centers if resource demands of virtual networks at different periods are given in advance. Otherwise, we develop a prediction algorithm to predict the periodic resource demands. We fourthly investigate cloudlet placement in a Wireless Metropolitan Area Network (WMAN), to enable pervasive cloud services for mobile users. We formulate a novel capacitated cloudlet placement problem that places K cloudlets to some strategic locations in the WMAN such that the average cloudlet access delay of mobile users is minimized. For this problem, we propose a fast yet efficient heuristic for it, and a novel approximation algorithm if cloudlets have identical processing capacities. We propose an efficient online algorithm for assigning user requests to cloudlets if the cloudlets are placed. We fifthly evaluate the performance of the proposed algorithms through experimental simulations by using real and synthetic datasets. Simulation results show that the proposed algorithms outperform the existing ones significantly. We finally conclude our work and discuss potential research topics.