Research On Resource Pool Scheduling Management Based On SDN / NFV

With the development and popularization of a series of new technologies and new services with the Internet of Things as the core,such as cloud computing,big data,and "Internet +",data resources need more stable,efficient,and flexible management algorithms in network information technology.So the research on resource scheduling under SDN / NFV network architecture came into being.This makes the network functions flexible and expandable in a virtualized environment,and each network component can be extended and integrated through virtual network functions,thereby providing a variety of network cost reduction services.It mainly uses the SDN / NFV technology and powerful machine learning tools to design resource scheduling algorithms based on the collected data information in the infrastructure virtualization environment built using virtualization technology.There are two main problems in this field.On the one hand,at the virtual infrastructure level,due to different workloads,virtual machine resource allocation is irrational,resource waste and resource competition are caused,and system performance decreases and interference factors increase.On the other hand,at the level of virtualized network function request,different service function chain requests are reasonably scheduled,so as to maximize the network function and resource allocation,and increase the programmability and automation of resource scheduling.Aiming at these two problems,this article conducts the following research:(1)Aiming at the scheduling of virtualized infrastructure resources,in order to reduce the interference caused by resource waste and resource competition,this paper uses a virtual machine scheduling algorithm based on workload classification of energy consumption and interference perception.That is,the prediction ability of the machine learning model is used to classify the workload of the virtual machine,and finally a more optimized integration scheme is reached.This article uses workload data from Microsoft Azure,uses the predictive capabilities of Artificial Neural Networks(ANN)to classify the workload of virtual machines,and introduces a classification and energy-aware VM scheduling algorithm(Classification based Energy and Interference Aware Algorithm(CEIAA)to solve this problem.This algorithm greatly improves the energy efficiency of the system and reduces the occurrence of Service Level Agreement Violation(SLAV).(2)Aiming at the problem of service function chain scheduling in a virtualized network environment,this paper describes it as a 0-1 integer programming(Binary Integer Programming,BIP)service function chain scheduling model.Combining with the Deep Reinforcement Learning(DRL)algorithm,we propose a DDQN model-based SFC scheduling algorithm(Double Deep Q Network Service Function Chain Scheduling Algorithm(DDQN-SFCSA))to optimize the high-dimensional solution space,and obtain Small solution space and determine the optimal solution,and placement and release of virtual network functions(VNF)based on a threshold strategy.This article uses Google Cluster-trace Data to perform simulation experiments,which shows that the algorithm is in terms of the number of rejections and rejection rates of service function chain requests,throughput,end-to-end delay,VNF instance runtime and load balancing.Can improve network performance.