Non-Cooperative Gaming And Bidding Model Based Resource Allocation In Virtual Machine Environment

Nowadays with the rapid development of the information technology, the performance of a single computer is improved continuously. However, a single computer’s resource utilization is still very low. Virtualization technology can dynamically manage resources, and isolate the software systems from specific hardware architecture, which can efficiently improve the utilization of computing resources, such as processor, memory. Therefore, virtualization technology has been widely used, such as data center server consolidation, storage virtualization and cloud storage. In the virtualization system, multiple guest virtual machines share the same hardware resources, which can effectively improve the utilization of system resources. In this paper, we study the system-level virtualization and research how to do resource management strategies. However, different types of service applications have different demands on the mount of resource in practice. Especially in high concurrency situations, it will be more competitive for guest virtual machine to get hardware resource. So the resource should be allocated based on the workload of the guest. The key problem is how to allocate resource to each virtual machine (VM) reasonably and efficiently, offer differentiated services and assure the service quality at the same time in a highly competitive environment. Based on the context of system-level virtualization, we study the resource management strategies of virtualization.In this paper, the main contents and innovations of this study are summarized as follows:(1) A new resource allocation scheme based on economic theory is proposed. In this paper, we consider the virtual machine resource as commodity and VMs as customers, respectively. The only way for a virtual machine to get the resource is to bidding. If a VM bids higher, it will get more resource. The price a VM bids reflects it’s level of urgency for resource.(2) A reasonable utility function based on biding model and non-cooperation game model is proposed. Although no good definition of zero in the utility function, in this paper we give the proof of uniqueness and optimality of optimal solution in each virtual machine.(3) According to the utility function, we construct a Nikaido-Isoda function and prove it is a weakly convex and concave function. Also we prove that our utility function could converge to the Nash equilibrium point by using relaxation algorithm.(4) In order to verify the validity of our model, we build Web system environment based on the TPC-W benchmark source code. The system contains books selling online system, load generator which impersonates users’requests, CPU utilization collector and game collector. We use load generator and game collector to allocate virtualized resource, and we make a comparative experiment to compare the average response time and resource utilization without game collector. We conclude that the resource utilization is significantly improved, and the cost of the game collector is very small.In multiple virtual machine environments, the virtualized resource allocation scheme based on game theory, on one hand, can effectively improve the utilization of system resources and ensure fair distribution of the resources. On the other hand, if the resource is limited, our scheme can make a differentiated distribution to ensure the service quality.