| The key characteristics of cloud computing is to make IT resources can be leased on demand like water, electricity and gas. The realization of the lease can't depart from the support of virtualization technology. The main method is to gather physical resources together and to establish a shared virtual resource pool by virtualization, and then virtual resources leasing can be achieved dynamically. In cloud environment, IT market is divided into two parts:infrastructure provider (InP) and service provider (SP). InP owns and maintains hardware resource, and abstracts it to construct virtual resource pool. SPs rent the virtual resource on demand to build their custom virtual networks. Such typical multi-tenancy mechanism in cloud computing is great significant for enterprises to save cost and improve the utilization rate of resources.As a result, efficient, reasonable leasing of virtual resource is the key research problem of cloud computing. The existing virtual resource management research are mainly concentrated in the fields of leasing transaction mechanisms and virtual network deployment. The Leasing transaction mechanism is more macro, it aims at the market competition status of supply and demand relationship between InPs and SPs in multi-tenant cloud environment. The goal is to maximize the interests of the whole society and to ensure fair and effective competition environment. Virtual resources deployment is to study how to satisfy more users as much as possible in limited hardware resources with the hardware resources utilization efficiency as the goal. It allocates resources and create a virtual network according users demand directly, which is belong to the virtual network mapping problem (VNE).Although there are many virtual resource allocation algorithms have been proposed, but the existing research still have the following problems:(1) the resource allocation pricing mechanism is not flexible; (2) Resources utility of the underlying network is low; (3) Mapping success rate of virtual networks is low; (4) The physical network is prone to resource bottleneck; (5) The virtual network request's parameters is fixed; (6) Reliability of the virtual network is lack. Therefore, in view of the above problems, this paper carried on the thorough research to the virtual resource allocation problem, and obtain the following results:(1) Because of the competitive relationship between SPs in cloud market, they may not fully share information such as the rent price. So we use Hidden Markova Model (HMM) to forecast current rent price of SPs according to their historical resources demand. Then build a dynamic pricing game model to encourage SPs to choose the optimal bidding strategy so as to realize benefit maximization. In the resource allocation phase, resource allocation model with variety resource unit price as benchmark is designed. The model supports simultaneous allocation for multiple SPs and multifarious resource type. It also can increase the benefit of InP and improve the fairness of competition in the resource market.(2) For virtual network mapping problem under multi-tenancy market, in order to reduce the average underlying link load and speed up the mapping efficiency, improve the utilization efficiency of the underlying physical resources, this proposal introduces a virtual node mapping rules with discrete particle swarm optimization (DPSO), which named MLB-VNE-SDPSO algorithm with physical nodes reusable and load controlled embedding. The algorithm can save physical link bandwidth resources and reduce the complexity of link mapping process. It also can guarantee the network load, get a good physical nodes utilization and improve the benefit-cost ratio of virtual network. In order to further improve the previous algorithms on the efficiency of large scale network, the crossover operator is introduced into MLB-VNE-SDPSO algorithm, and a hybrid intelligent algorithm is designed. The algorithm can solve the problem that the traditional particle swarm algorithm easy to fall into local optimal point and cannot reach the global optimal. It can also help the physical network to obtain higher benefit-cost ratio.(3) For the reason that topology of the virtual network have an important impact on the mapping success rate, we redefined node's comprehensive ability, put forward a kind of virtual network mapping algorithm based on topology awareness (WD-VNE algorithm). In the mapping phase, a topological perception and measure method include node connection degree and the comprehensive ability to gain optimal embedding solution. A sliding window technology is introduced into the algorithm for pretreatment of virtual network request. Simulation results show that the algorithm can gain good accept ratio and benefit-cost ratio.(4) In order to improve the acceptance rate and utilization of the underlying network, also avoid bottleneck of node and link in the underlying network, a multi-objective optimization mathematical model is introduced for multi-objective optimization mathematical model problem. A heuristic virtual network reconfiguration algorithm based on meta-heuristics mathematical method is put forward, which migrates running virtual networks according to the optimal solutions gain by the heuristic algorithm. The algorithm can significantly reduce the maximum load of physical nodes and links, effectively avoid bottleneck on the physical network and guarantee acceptance rate of virtual network request(5) For the reason that the resources parameters of the virtual network request is dynamic in practice, a virtual network mapping algorithm (VNE-DR) based on dynamic virtual network requests is puts forward. On the basis of mixed linear programming theory, an embedding model with multi object as minimum mapping and migration cost is established. The algorithm adopts multiple queues to store different types of virtual network request. Priority will be given to those requests which need to release resource to get more idle resources for supporting more virtual networks. The algorithm can reduce mapping and migration cost.(6) In view of fault tolerance, due to the physical network node and link failures, a reliable embedding algorithm is presented which first add backup virtual nodes and links for virtual network and then go to mapping phase. An integer linear programming based model is established with the object of minimizing the mapping cost, which can deal with node and link failures in physical network and guarantee fault-free running of the virtual network. The algorithm determines which nodes and links to be back up by evaluating the virtual network nodes' importance degree and key structure of the topology. Then extend the virtual network topology by setting up an augmented topology with additional backup resources and embed the augmented topology. The algorithm can improve reliability of the virtual network.In summary, this dissertation dedicates two perspectives of virtual resource allocation and virtual network mapping in multi-tenancy environment under cloud computing, studies virtual resource allocation algorithm based on incomplete information game and virtual network embedding algorithms suitable for different scenarios. Lots of theoretical analysis and experimental results show prove the effectiveness and efficiency of these methods. We hope that these approaches and techniques could make contributions to cloud resource management and scheduling systems. |
PDF Full Text Request