Game Theory Based Resource Scheduling Approaches Of Cloud Computing

By the emergence of cloud computing, physical resources in datacenter are virtualized and supplied to deal with the dynamic resource requirements of customers. Due to the continuously growing scale, heterogeneity and complexity of cloud datacenter, it is challenging to design an efficiency resource scheduling mechanism for a cloud environment.For cloud providers and cloud customers, these two parties both pursue the maximization of their own interests. Most of the current studies focus on only one party’s benefit. Howerver, an agreement should be reached to satisfy the two parties, otherwise, cloud providers can refuse to provide resources or customers can choose other providers. Thus, the resource scheduling mechanism should be considered to ensure the benefits of both cloud customers and cloud providers.Game theory is a mathematical theory for rational decision making problems, which has been widely applied to cloud resource management problem recently. This thesis focuses on the resource scheduling problem in cloud computing and the analysis of optimal objectives for different cloud participants. Game theoretic methods are applied to model the various resource management scenarios and design the optimization algorithms associated with the multiple objectives.The main contributions of this thesis can be summarized into the following four aspects:(1) A uniform descrption is proposed to define cloud resources, resource prices and customers’ jobs. In the light of the different optimization objectives, two resource scheduling games are modeling for cloud customers and cloud providers. Each cloud customer is modeled as a player in the former game, and each physical machine with available resources is modeled as a game player in the latter one. These two game models provide foundation for the design of game theoretic resource scheduling approaches under different cloud scenarios.(2) A finite extensive game with perfect information is modeled, and a resource scheduling algorithm FUTG based on backward induction is also proposed to address the issues of unfair resource distribution among multiple customers and the low resource utilization in a Hypervisor virtualized cloud datacenter. This algorithm can significantly improve the efficiency of resource usage by reducing the skewness on each physical machine and balancing the ulitization among multiple physical machines, as well as improve the fairness among cloud customers.(3) A two-phase resource scheduling approach is proposed for a container virtualized cloud datacenter. In the first phase, the queuing theory is applied to calculate the optimized amount of containers for each customer, according to the maximal response time and the cost in QoS constraints. In the second phase, the containers to be created and available physical machines are modeled as the two sides of stable matching problem, and a container stable placement (CSP) algorithm is also proposed to achieve a resource scheduling scheme. This approach reduces the response time of customers’ jobs and improves the resource ulitization rate of cloud datacenter, while guaranteeing the multi-QoS constraints.(4) An optimal cooperative cloud federation formation approach is proposed according to the stability and fairness principles of cloud federations. Moreover, to model the resource scheduling problem under multiple cloud providers, a cooperative game based resource scheduling algorithm is presented to optimal the QoS of cloud customers and the average resource utilization of the cloud federation.