Research On Energy Efficient Virtual Network Embedding Algorithm For Data Centers

Cloud computing has changed the business model of the Information Technique(IT)industry.As the core infrastructure of cloud computing,large-scale Data Center(DC)ensure that users can access computing,network,and storage resources anytime and anywhere on demand.However,while facing significant development opportunities,data centers also face the challenge of high energy consumption,which not only increases the operating costs of cloud service providers,but also seriously restricts the development of data centers.Therefore,it is urgent to deeply research Network Virtualization(NV)resource embedding technology in data centers.The key to realize Network virtualization is Virtual Network Embedding(VNE).By using Virtual Network Embedding technology,more virtual resources can be integrated into fewer physical resources to achieve the goal of high energy efficiency.However,most of the traditional VNE algorithms focus on the study of Virtual Network Request(VNR)acceptance ratio,cost and resource utilization,and often ignore the important index of energy consumption.In addition,existing VNE algorithms related to energy saving are often constrained by load balancing,congestion,delay and other indicators,which restrict the effect of energy saving.In addition,the joint node/link embedding method makes the algorithm need too large search space and too high complexity.Considering the defect as above,the thesis endeavors to research on the high energy efficient virtual network embedding algorithm of data centers.Firstly,this thesis proposes an efficient virtual network embedding algorithm based on packing algorithm for computing common task scenarios.To be specific,a virtual network embedding system model is presented.With the constraints of server CPU cores and bandwidth resources,an optimization problem aiming at minimizing energy consumption is established.Secondly,because the problem is NP-Hard,this thesis investigates the low complexity heuristic algorithms.In the virtual node embedding stage,the First Fit(FF)algorithm and the Best Fit(BF)algorithmare respectively used to embed the virtual nodes.In the virtual link embedding stage,the shortest path algorithm is used to map the virtual links.Finally,the simulation results show that the proposed high-efficiency virtual network embedding algorithm based on the packing algorithm has better energy saving effect than the adaptive algorithm while keeping the cost of cloud service providers and the utilization rate of server resources unchanged.Secondly,an efficient virtual network embedding algorithm based on Hypergraph Matching(HM)is proposed for computationally intensive task scenarios.The system model is transformed into a weighted hypergraph model.For this model,each vertex in the hypergraph represents a virtual node,and each hyperedge represents the same physical network server embedded by one or more virtual nodes.The weight of the hyperedge is defined as the negative value of the accumulated energy consumed by the virtual node embedding to the server.Thus,the problem of minimum energy consumption of virtual network embedding becomes a problem of finding a perfect matching with maximum total weight in the hypergraph.Since the problem is NP-Hard,this thesis investigates the low complexity heuristic algorithms.Then,the hypergraph is transformed into a conflict graph,and the suboptimal solution is obtained by finding the maximum weight value of the independent set of vertices whose edges do not intersect.Then,the independent set is checked to ensure that all virtual network requests are embedded successfully.Finally,verified by simulation in computationally intensive task scenarios,in keeping the cost of cloud service providers and server resource utilization is constant at the same time,the energy-saving effect of the proposed algorithm is superior to other algorithms.