Optimization Of Resource Configuration For Cloud Computing Based On Game Theoretical Methods

Due to the advantages of high computing power,high scalability,low service cost and accessibility,the cloud computing is becoming the platform of choice for a number of applications.With the increase and evolution of user requirements,the scale of the cloud computing is also growing,and the service model is changing.Whereas,the energy consumption in data centers is soaring,and the resource management for clouds becomes more complex.Being similar with public utilities such as water and electricity,the cloud provider aims to obtain profit by offering computing resources to customers.Accordingly,to effectively enhance the quality of service for cloud users while promising the improvement of cloud providers' profit,optimizing resource management for the cloud computing has become a fundamental and unavoidable issue.In this thesis,we focus on the resource management optimization from four aspects,which are energy saving,load balancing,pricing strategy and resource provisioning.Considering the difference of the optimization goal for processing performance,energy consumption in various scenarios and the problem for pricing and allocating the multi-type virtual machines in cloud computing,we utilize game theory to propose several effective strategies and algorithms for them,in which the economic benefits of cloud providers and the satisfaction of cloud users are taken into account.The major work and contributes of this paper are listed as follows.First,this thesis focuses on the problem of the tradeoff between the performance and energy cost in cloud data centers.We model the load balancing problem in clouds as a cooperative game among the multi-server systems.The objective is to minimize the energy consumption of cloud data centers while promising an acceptable average response time for their customers.In our model,the multi-server systems,who act as players,collaborate to reach the Nash bargaining solution(NBS)of load balancing problem in clouds,such that the optimal global performance is achieved in the game.The M/M/c queuing model is employed to formulate and explore the problem of optimal load distribution for multiple heterogeneous multi-server systems in a cloud computing environment.In our methods,the computing resources are provided in a greedy manner,and then the users' tasks are allocated along with the concept of the NBS.Given NBS is able to deliver the optimal global solution,our methods can achieve better processing performance at the same energy consumption.Simulated results demonstrate that the proposed technique has immense potential to balance the tradeoff between the energy saving and performance for the cloud data centers.Next,We employ game theoretic approaches to model the problem of minimizing energy consumption as a Stackelberg game.In our model,the system monitor,who plays the role of the leader,can maximize profit by adjusting resource provisioning,whereas scheduler agents,who act as followers,can select resources to obtain optimal performance.In addition,we model the problem of minimizing average response time of tasks as a noncooperative game among decentralized scheduler agents as they compete with one another in the sharing resources.Corresponding algorithms are presented to implement the game models.In addition,the size of users requirements fluctuates periodically over time.To address this problem,We convert the making decision problem of the leader into an alternative problem to obtain the list of the optimal resource configurations for various load rate conditions in the data center.According to our approaches,the provider only aims to retrieve the right configurations from the stored list depending on the current load rate,rather than solve the problem directly again during runtime.Simulation results demonstrate that the proposed technique has immense potential to improve energy efficiency under dynamic work scenarios without compromising service level agreements.Then,we focus on the resource management optimization for edge cloud computing,and model the resource competition among cloud users as a non-cooperative game,where the objective of the cloud user,who acts as the player,is always to search suitable allocation strategy to minimize the expected response time for her tasks.In our schemes,we first construct the utility function of users with queuing model,and prove the existence of Nash equilibrium for the formulated game.Then,using the concept of Nash bargaining solution to calculate optimal task offloading schemes for cloud users,we propose a low-complexity distributed task assignment algorithm.The results of simulated experiments demonstrate that our method can quickly reach the Nash equilibrium point,and deliver a satisfying performance at the expected response time of the user's tasks.Finally,we investigate the problem of pricing and allocating the multi-type virtual machines in cloud computing from the perspective of game mechanism,and propose a combinatorial auction-based mechanism to address such problem.Moreover,the multiple criteria decision making technique is adopted in order to fairly measure the bundles of VMs requested by distinct users with different requirements.The proposed mechanism combines two general ideas: consensus estimate that can avoid market manipulation and yields an approximate optimal target revenue with the consensus estimate technology,and RevenueExtraction that can determine the winners and equally shares the target revenue generated by consensus estimate among them with a single sale price.Using the two ideas,the proposed mechanism can simultaneously promise truthfulness and envy-freeness while achieving an approximate optimal revenue.Additionally,our schemes can provide computationally efficient solutions being suitable for execution in short time window auctions,and deliver stable and desirable performance,especially in large-scale cloud markets.