Research On Resource Allocation Based On Game Theory In Service Cloud

As the internet, especially mobile internet, becomes more and more popular, the number of users and applications increases rapidly. In the context of numerous users and highly dynamic environment, traditional service platforms will not satisfy the quality of service, which lead to the concept of cloud computing. The reliability, availability and elastic of cloud computing guarantee the high qualities of users’services as well as low cost of providers.With the pay-as-you-go payment method, cloud computing reduces the expenses of users. Considering the advantages of cloud computing, research institutions and enterprises reform platforms as well as applications based on cloud computing. Nevertheless, the cloud resource allocation is a challenging problem. The unreasonable resource allocation approach would result in low resource utilization and sometimes even difficult to meet the dynamic resource requirements of cloud services.The existing resource allocation algorithms optimal allocate cloud computing resource from the perspective of system. However, cloud computing is a kind of utility computing, where providers and users have their own supply and demand strategies based on the complex competing among them. In this context, game theory models and approaches allow to gain an in-depth analytical understanding of the cloud resource allocation problem. Although lots of works solve cloud resource allocation problem based on game theory, there are still many challenges. There needs a fast and efficient algorithm to allocate cloud resource, which is scalable with respect to the number of resource types and users.The large scale of various cloud users cause methods such as multi-constrained conditions will not precisely describe different demands of users. In addition, it is unpractical that providers and users gain Nash Equilibrium by multistep negotiation in dynamic cloud environment. Finally, an effective cloud service deployment algorithm for interdependent cloud services is urgent.Relying on game theory, this dissertation launches the research in view of these limitations. The main contributions of this dissertation are as follows:1. VM allocation algorithm based on Dutch auction. The multi-instance Dutch auction is proposed as an efficient VM allocation algorithm, whose maximum duration is independent of the number of the bidders participating. Also, it is scalable with respect to the number and capacities of VM types and the number of users. Providers will gain optimal social welfare or revenue by determine the terminal time of the algorithm based on reserve price.2. Computing resource allocation algorithm based on price proportion. To efficient allocate divisible resources such as CPU and memory, we use the SLA-based utility function to characterize various services requested by users with different financial capacities and price proportion method to allocate resource. Furthermore, considering various scenarios, two cloud resource allocation schemes based on the algorithm are proposed.3. Bandwidth allocation algorithm based on greedy auction. Analyzing disadvantages of centralized and distributed mechanisms, the multi-bid greedy auction mechanism is proposed for cloud bandwidth allocation problem, which is applied to any types of demand (e.g. elastic, real-time, rate-adaptive and stepwise). Considering the scenario of elastic demand, the multi-dimensional bid greedy auction mechanism is proposed, who save the transmission energy.4. Service deployment algorithm based on congestion game. For complex and interdependent cloud services, the service deployment problem is modeled as a congestion game problem. Considering both efficiency and expense, it is achieved optimal cloud service deployment strategy based on congestion game model.To verify the performance of resource allocation and service deployment algorithms mentioned above, the dissertation analyze them both in theory and experiment. The results show that the algorithms proposed gain high level performance, which is also applied to the large scale and dynamic cloud environment.