Research On Resource Management And Scheduling Based On SDN/NFV

Since the 21st century,the rapid development of the Internet has become one of the important infrastructure supporting the operation of modern society.With the application of big data and artificial intelligence,existing networks have been unable to meet the requirements of new technologies on heterogeneous,dynamic,virtual and open networks.Software-defined Networking(SDN)and Network Function Virtualization(NFV)have become new technologies for reconstructing Network architecture and promoting Network innovation.It is recognized as the mainstream technology of next generation network by ITU telecommunication Standardization Department.SDN realizes the centralization of network control functions and can build multiple isolated networks on the common underlying network infrastructure.NFV decouples the network functions from the physical devices of network,that run network functions by using virtualization technologies to design,deploy,and manage on highvolume servers.SDN and NFV together can provide customized services for users on a general business platform,providing principled solutions to the increasing demand for existing network traffic,uninnovate network structure and high cost of network management.However,at present,the network topology and structure characteristics are complex,resource attributes and task types differ greatly,and the softwareization and openness of SDN and NFV results in problems,such as virtual network resource management and control of hybrid software-defined network,dynamic construction of service functions,interactive application fairness and heterogeneous resource allocation.These problems restrict the development and application of software-defined network and network function virtualization technology.In order to accelerate the application of SDN and NFV into the optimal and upgraded of information and communication network structure,improve the utilization rate of network resources and reduce operating costs,this paper summarizes and analyzes the achievements in the field of software-defined network and network function virtualization resource management and scheduling.In-depth research is carried out on mixed SDN resource embedding management,NFV resource scheduling and network function routing service view,SDN interactive application control and fair allocation of link resources,as well as NFV cloud heterogeneous resource allocation and management.The main research results are as follows:(1)Aiming at the low efficiency of network node upgradation and resource mapping in hybrid SDN multi-tenant virtual network management,a resource embedding service mechanism under hybrid SDN network management is proposed.The research firstly considers the optimal deployment of some SDN nodes in the hybrid SDN network,and then uses the characteristics of SDN to allocate resources to virtual network requests.Specifically,the management mode is put forward based on the improved PageRank SDN upgrades polymerization rich resources of the underlying network nodes.The feature matrix information of the network underlying structure including nodes and topology is included.And by adopting the mapping scheme including SDN forwarding nodes,the problem of resource waste caused by the large span of physical nodes can be reduced.In the case of resource conflict between virtual network request and existing service,the complex and inefficient problem of virtual network configuration is solved by using the low complexity of node output policy of hybrid SDN network.Simulation results show that the proposed strategy has good performance in the aspects of request acceptance rate,link and node pressure and conflict response time.(2)To solve the problems of lack of dynamic combination,high latency and low resource utilization in network function management and scheduling model,a virtual network function scheduling and resource optimization algorithm is proposed.Software-based network functions imposes significant performance overhead,and the latency of the service chain increases linearly with the length of the service chain.The algorithm is modeled to reduce the total service delay,and a centralized view construction mechanism of network function service chain is designed as the convenient solution.The service policy is transformed into high-performance network function service logic by identifying network function dependency,and the network topology information is considered while the network function is scheduled.The mechanism of path equivalence and network function parallelism is used to improve the reusability of network function.Simulation results show that this mechanism has good processing timeliness and low service delay.(3)Aiming at the problem of inequity among end users caused by lack of system coordination due to multi-interaction applications competing for network resources,a model of equitable distribution of network resources based on SDN is proposed to realize the joint allocation and scheduling of control and link resources.SDN expands the scope of network resources,and both controller and bandwidth become effective network resources.The allocation of control and link resources not only affects the ability to process network flows,but also affects the application network delay,thus bringing bad user experience.Therefore,on the issues such as controller processing time and link congestion that affect the fair experience of interactive applications,this paper studies the joint consideration of control and resource scheduling and fair allocation in links in combination with the network context.An adaptive scheduling method is proposed to dynamically adjust the weight of bandwidth resource allocation according to link usage,which realizes the active adjustment of controller and the sharing of link scarce resources.The final results show that the fairness between user flows is guaranteed under the premise of ensuring network efficiency.(4)Aiming at the problems of complex resource types and lack of dynamic deployment and unified management of network functional resources for applications in the new NFV cloud network architecture,a collaborative and complementary resource allocation strategy based on prediction was designed.The research firstly extracts the network feature information vector,then fuses the history and future mapping sequence and task type in the model.By both short-term and long-term memory network learning objectives from the original stream data flow estimation model,the characteristics of a flow network is deduced function classification and processing power requirements.Based on this,a dynamic network function extension and collaborative and complementary resource allocation algorithm are designed to map different types of resources and services into unified features.By using the joint scheduling scheme of network function deployment and resource allocation based on machine learning,the heterogeneous resource allocation problem is mapped to a multidimensional and active planning problem,which reduces the fragmentation of virtual machine resources allocation due to the cloud-based workload heterogeneity and request dynamics,and improved the utilization of server resources.