Resource Allocation Of Clustered Virtual Network With SDN In Cloud Computing

As cloud computing becomes more and more massive and dynamic,it is common for cloud providers to manage the entire cloud computing resources while meeting the resource requirements of users' applications.The goal of cloud computing is that tenants can rent cloud resources according to the pay-as-you-go model,and cloud providers can manage cloud resources while ensuring the scalability and effectiveness of services.The emergence of network virtualization technology has brought an opportunity for the efficient management of cloud computing resources,which "abstracts" the entire cloud data center(physical infrastructure)into a huge virtual "resource pool" to meet the needs of multi-user sharing of resources simultaneously.That greatly reduces the difficulty of cloud resource management for cloud providers as well as management and maintenance costs.Software-defined networking(SDN)that represents the direction of future network has a global sight of the whole network,which fine-tunes the entire cloud data center resources and makes resource allocation more serviceable and effective.However,due to the large number of tenant services and applications usually share the same resource pool,making the data center environment very complex,it is a great challenge to effectively allocate resources to meet the requirements of multiple users while considering the performance of the whole network.To address this problem,we first construct a network model of virtual resource allocation based on software-defined networking of fat-tree data center in this paper.Then,the paper proposes a clustered virtual network(CVN)abstraction strategy and a dynamical heapsort algorithm to find the pod with the most appropriate number of virtual machines(VMs)and construct a CVN time-efficiently by utilizing resources within one pod.Subsequently,we construct a crossing pod CVN(CP-CVN)if there are not enough resources in any one pod,and propose a max-min strategy to allocate VM resources from different pods.Furthermore,we distribute network resources for the construction of CP-CVN based on the VM allocation,and the network optimization problem is formulated as a Linear Programming(LP)problem aiming to maximize network throughput.To achieve computation feasibility in massive data center networks,we propose an approximation primal-dual algorithm for solving this LP problem.Theoretical analysis shows that the proposed primal-dual algorithm is feasible and suitable for solving massive computation problem and guarantees network performance in SDN.Finally,the simulation experiment is emulated through the network simulator Mininet.Mininet is a network emulator,which uses lightweight virtualization to deploy large-scale networks on a single system,such as a computer or virtual machine.We design the simulation experiment from multiple dimensions,such as network topology,communication pattern,resource allocation and flow generation.The experimental results show that the CVN algorithms proposed in this paper outperform Cloud Mirror,Oktopus and MILPFlow algorithms in terms of network throughput and average delivery delay,which verifies the correctness of our theoretical analysis.