Research And Implementation Of Resource Management For Derterministic Applications

With the development of science and technology,more and more new applications appear,such as Augmented Reality/Virtual Reality,holographic communication,process monitoring,etc.These applications have ultra-low latency requirements,and some of them also have deterministic latency requirements.This poses a huge challenge to future networks.Although some organizations have begun to study deterministic networks,such as IEEE Time Sensitive Network Working Group,IETF Deterministic Network Working Group,Network 5.0 Industry and Technology Innovation Alliance.However,the current research is focused more on the problem of ultra-low latency data transmission,than researching on the deterministic latency flow.In addition,many research focuses on the delay guarantee of the second and third layer routing in the OSI model.There is little analysis from the perspective of resource management.Therefore,this thesis focuses on the resource management of deterministic applications.This thesis investigates and analyzes the requirements and characteristics of future network applications.According to the requirements of applications,future network applications are divided into four categories.Aiming at the resource allocation and routing decision of these four kinds of applications,this thesis discusses a resource management framework based on SDN,and defines the functions of each plane.On the basis of the framework,the thesis studies the resource allocation problem and data flow classification and identification problem for deterministic applications.The main contents include:(1)The thesis discusses the problem of data forwarding and resource allocation for deterministic applications in Computing Power Network,and proposes a resource allocation method for deterministic applications based on two-layer particle swarm optimization.This method starts from the resource distribution network in the Computing Power Network and considers the bounded latency requirement.The multi-dimensional resource(computing,storage and bandwidth)allocation problem is transformed into a multi-constraint optimization problem,which is solved by a two-layer particle swarm optimization algorithm.The purpose of the method is to improve the network request reception rate,throughput and resource utilization.Simulations are made to compare the effects of different environmental parameters on network resource allocation,and the experimental results shows that by considering the bounded latency requirement,the network request receiving rate,throughput and resource utilization can be improved(2)This thesis classifies the future network applications,and uses different resource allocation and path selection methods for each type of application to improve the network request reception rate and throughput.Therefore,it is necessary to classify and identify data flows,which is the premise of resource allocation and routing.In order to classify data flow more accurately,the thesis implements a data flow classification method based on hybrid clustering.This method combines the feature of traditional clustering methods,and the experimental results show that the classification result is more reasonable than the traditional clustering algorithm.In order to improve the accuracy of data flow identification,this paper introduces the idea of residual into deep fully connected neural network,and realizes an improved data flow identification method based on residual.Simulation results show that this method has high identification accuracy.(3)The thesis designs and implements the resource management framework for deterministic applications,and describes the functions of each module in the framework and the information interaction process between modules in detail.