Research On Fault Detection Technology For SDN Based On Knowledge Graph

With the development of computer network in the new era,all walks of life have begun digital transformation,and users have put forward higher requirements for the reliable operation of the network.Fail to detect faults in time will have an impact on daily life and social production.The distributed architecture of SDN brings flexibility to the network,but also brings new fault problems,which traditional detection methods are not necessarily applicable to.At the same time,the new technology of knowledge graph has been put into use in medicine,e-commerce and other fields with excellent knowledge representation and reasoning ability.It uses the form of triples to describe the entities in the objective world and the relations between entities.Therefore,thesis combines SDN with knowledge graph,and studies the scheme of using knowledge graph to realize SDN fault detection.The main work includes:First,the thesis designs an SDN fault detection framework based on knowledge graph,including knowledge storage,knowledge management,and fault detection application.Knowledge management regularly updates the knowledge graph in the storage,and provides interfaces for knowledge retrieval and knowledge reasoning for fault detection applications.In this way,administrators only need to focus on the application logic and not the underlying logic.Second,we construct the knowledge graph scheme level for SDN fault detection,and use the real-time topology information of SDN network as the data level.Then using the centralized and unified characteristics of the controller and the ability of knowledge graph to represent entity relationships,associates entities with complex relations such as SDN network devices,ports,and flow tables to provide a basis for fault detection.Third,fault detection methods based on abnormal traffic and flow rule conflict are proposed.For fault detection of abnormal traffic,the number of Packet-In packet is selected as the abnormal monitoring index of the switches,and then the source address entropy value is calculated by filtering the flow rules to realize the abnormal monitoring of the host.After that,we use knowledge graph to reason about topology information to find out the source of abnormality;For the fault detection of flow rule conflict,the multibranch matching relation classification method is converted into the form of production rules,and the flow rule conflict detection in single flow table and among multi flow tables is realized by using the symmetry and transitivity of attributes.Finally,we choose the ONOS controller and the Mininet network simulation platform to build the environment,use the Jena tool to regularly update the network information on data level of the knowledge graph,and perform functional and performance tests.The experimental results show that the knowledge graph can complete the fault detection based on abnormal traffic and flow rule conflicts,the writing and processing logic of the rules are clear,the good knowledge representation makes the system readable and easy to modify and expand,and it is convenient to build a unified management system.