Research On Dynamic Resource Allocation Strategies And Algorithms Of Virtual Networks

The main function of network virtualization technology is to virtualize the resources leased by multiple infrastructure providers and allocate corresponding network resources for various Internet services.Virtual network embedding(VNE),as the key to virtual network technology,solves how to efficiently map multiple heterogeneous virtual networks(VN)to the underlying physical network,which proved to be a typical NP-hard problem.The existing VNE research has many defects.First of all,most VNE methods are limited to design static VNE algorithm,that is,the resource requirements of nodes and links of each virtual network request(VNR)from network users will not change during its entire life cycle.However,in reality,the resource requirements of users change dynamically over time.Therefore,assigning a fixed amount of network resources to users according to their peak resource requirements will result in a lot of waste of resources.Second,traditional resource pricing methods mostly use static symmetric pricing,that is,users and service providers sign long-term fixed bandwidth lease contracts,which easily lead to unbalanced use of network node and link resources.Finally,with the rapid increase in the number of internet users,network resources have become increasingly scarce,and theories and mechanisms related to resource sharing urgently need to be further improved.This paper focuses on network modeling,resource allocation,resource pricing and resource sharing in dynamic virtual network mapping scenarios.The specific research content is as follows:(1)Aiming at the network mapping problem of dynamic changes in virtual network resource requirements,a dynamic virtual network mapping algorithm based on group search optimization algorithm and RBF neural network is proposed(GSO-RBFDM).First,perform network modeling according to the characteristics of the dynamic resource requirements of virtual requests,and then use the group search optimization(GSO)algorithm and the radial basis function(RBF)neural network algorithm to optimize the node mapping scheme and predict the users' dynamic resource requirements.Finally,according to the prediction results of RBF neural network algorithm,network resources are dynamically allocated for users.Simulation experiments show that the performance of the GSO-RBFDM is better than the existing static resource allocation algorithm.(2)Aiming at the large amount of calculation and long training time caused by the use of many instances as hidden units in the prediction process of the RBF neural network algorithm,a dynamic virtual network mapping algorithm based on the incremental design method of the RBF neural network is proposed(GSO-INC-RBFDM).First,improve the original RBF neural network algorithm,construct and train the RBF network structure through an incremental design method,namely INC-RBF,and then combine the INC-RBF and GSO algorithms,use the GSO algorithm to optimize the node mapping scheme and the INC-RBF algorithm to predict user dynamic resource requirements;Finally,dynamically allocate network resources for users according to the INC-RBF prediction results.Simulation experiments show that GSO-INC-RBFDM can not only reduce the amount of calculation and shorten the prediction time,but also improve the accuracy of resource prediction,thereby reducing the redundancy of resource allocation.(3)Aiming at the resource imbalance problem caused by the static symmetric pricing mechanism in the dynamic virtual network mapping process,a dynamic virtual network mapping algorithm based on resource dynamic pricing and genetic algorithm path optimization is proposed(GSGA-RBFDM).First,the physical node and link resource is dynamically priced,that is,the price of a unit resource changes dynamically with the remaining quantity.Those nodes or links with sufficient remaining resources and low unit price will be preferentially selected as the mapping target,which reduces the number of bottleneck nodes and links and makes resource usage more balanced.Then considering the cost of nodes on the physical mapping path,the problem of finding a low-cost physical mapping path becomes a complex optimization problem.Genetic algorithm(GA)is widely used in combinatorial optimization problems.Therefore,it is feasible to apply GA to the mapping path optimization problem and find low-cost physical mapping paths through operators such as crossover,recombination and iteration.Finally,the RBF neural network algorithm is used to predict user resource requirements and dynamically allocate physical resources for them.Simulation experiments show that the GSGA-RBFDM can alleviate the problem of unbalanced use of resources and improve the network acceptance rate.(4)Aiming at the problems of node storage resource attributes and resource sharing that are often ignored in the process of dynamic virtual network mapping,a dynamic virtual network mapping algorithmbased on resource three-dimensional constraints and resource sharing is proposed(NMA-PRS-VNE).First,a network model based on three dimensional constraints of node computing resources,storage resources and link bandwidth resources is established.Then the users' dynamic resource requirements are divided into basic sub-requirements and variable sub-requirements.Just as the former represents the monopoly requirement for resources by VNRs,the latter represents the resources that multiple VNRs can share with the probability of occupying resources.Finally,calculate the resource usage collision probability between variable sub-requirements from multiple VNRs events,and compare with the reset maximum collision threshold to determine whether the same network resource can be shared.Simulation experiments show that the network mapping framework based on node multi-attribute constraints is closer to the actual network,and the NMA-PRS-VNE proposed based on this framework reflects better performance in terms of acceptance rate,network cost and average network revenue.