Research On Virtual Network Dynamic And Evolutive Management Technologies In Cloud Environment

Network virtualization technologies are decoupling network resources layer from physical layer, shielding network complexities for upper-class by aggregating distributed heterogeneous web resources, and upgrading the network management flexibility. Cloud computing combines with network virtualization technology makes more agile, efficient management of network resources possible. However, due to the dynamics of cloud applications, virtual network dynamic management and lacking of business aware capabilities, virtual network resource utilization ratio remains low in the cloud environment, and network management issues such as fragmentation is becoming more serious. Lacking ability to dynamically adapt to business demands of cloud application will make virtual network not being able to improve cloud application flexibility, agility, and resource utilization, but also the complexity of the cloud infrastructure will be increased, and result in potential risks about stability and performance for cloud applications and greater waste of resources causing by the extra built virtual networks layer. With the emerging and fast evolution of future-oriented "Terminal-Network-Cloud" models, the contradiction between lacking of ability to identify and continuing evolution to adapt to dynamic of cloud application, and the fast-growing connected terminals has been increasingly prominent. Currently, although there have been some achievements emerged in this research field, but still exists problems such as:lacking application service level objectives aware virtual network dynamic evolution management solution in cloud environment; business sensitive virtual network resources scheduling and topology structure reconfiguration; difficult in cloud application runtime relied virtual network risk analysis and the root cause positioning pending to be resolved. This thesis analyses existing problems and challenges, researches on current research status, and propose BDI (Belief-Desire-Intention) model based autonomous virtual resource management framework. Supported by this framework, we resolve the existing cloud application's service level objectives sensitive virtual network dynamic and evolutive management relevant problems by using split clustering of graph theory, Bayesian belief networks theory based probabilistic inference and linear programming modeling theory. The efficiency of existing virtual network reconfiguration algorithm is being improved also. Finally, based on theoretical study achievements, propose a design and implementation of virtual network management for cloud applications platform system to verify the technical feasibility. The main works of this thesis can be summarized as follows:(1) A Belief-Desire-Intention model and the multi-agent system based autonomic cloud application management framework design is being proposed, by the strategy of decoupling autonomous control functions from cloud application itself, the implementation complexity and difficulties of network access configuration and management in cloud environment are greatly simplified. The problems of reducing loss of network virtualization and virtual node migration effectiveness judgement being resolved also, and provide a framework for supporting dynamic evolutive virtual network management.(2) Propose graph theory split clustering algorithm based virtual network resources scheduling approach, by determining the virtual network bandwidth utilization of cloud-based applications described in cloud application container model, schedule the virtual resource for optimizing their network bandwidth resource allocation policy, easing network fragmentation and enhancing the network resources utilization.(3) Propose method for virtual network risk analysis, based on Bayesian belief networks propose uncertainty, which is caused by the dynamicity and diversity in computing and communication resource, causal-effect relation modeling solution for virtual network performance in dynamic heterogeneous network computing environment, and design strategy of using pre-given service level objectives to trigger Bayesian network MPE (Most Probability Explanation) probabilistic reasoning, hence judge if virtual network is in over/under provisioning state, realize automatic virtual network risk analysis and reconfiguration request generation process.(4) A virtual network reconfiguration method of combining virtual nodes/virtual link migration and replacement strategy is proposed, design solution for leverage the degree of satisfaction of virtual resource increase request demand by the virtual network reconfiguration ratio indicator, and by modeling the target problems based on linear programming theory, the virtual network reconfiguration algorithm efficiency is being enhance further.(5) Proposes design solution of an agile cloud application management platform, which with the capabilities of on-demand virtual network resources provisioning, risk forecasting, positioning and handling, and managing the complexity of the virtual network topology, provides the basic management platform level support for the "Terminal-Network-Cloud" architecture systems.