Research On Health Management Technology Of Publish/Subscribe Distributed System

Application system security is very important in national defense,military,industrial automation and other key fields.DDS uses a Qo S policy to define the information model,it helps system designers effectively build distributed applications based on the needs of specific data segments by providing standardized interfaces.The result is a data-centric publish/subscribe communication specification.Distributed publish/subscribe system based on DDS middleware has been paid more and more attention in military,industrial and other security critical fields.DDS publish/subscribe distributed system has been widely used in military,aerospace and other security key fields.The scale of hardware and software of the system is constantly expanding,and the composition and interactive logic are increasingly complex,any minor fault in the system will cause fault propagation and bring huge security risk to the system.Therefore,it is urgent to study the corresponding methods and mechanisms to ensure the availability,reliability and security of the system during operation.In view of the above problems,this thesis applies the software health management technology to the operation and maintenance of publish/subscribe distributed system,and puts forward the fault diagnosis scheme of publish/subscribe distributed system: firstly,a component fault pattern recognition method based on fuzzy clustering is proposed to identify faults in the system by component as a unit;then,a fuzzy Signed Directed Graph(SDG)model is proposed to describe the fault propagation mechanism in distributed publish/subscribe systems;finally,the fault root cause diagnosis method based on the fuzzy layered SDG model is proposed,and the fault root causes of the system are inferred from the identified faults.The main contributions of this thesis are as follows:(1)In view of the characteristics of diversified fault modes and fuzzy pattern boundaries in distributed publish/subscribe systems,a component fault pattern recognition method based on fuzzy K-means clustering is proposed.Fault pattern recognition consists of two parts:operation pattern mining and fault pattern recognition.Abstract models of normal modes and common fault modes are obtained from operation mode mining,and fault pattern recognition is realized based on the abstract model.(2)A fuzzy SDG model is proposed to describe the mechanism of system fault propagation in distributed publish/subscribe system,which is characterized by fault propagation among components through publish/subscribe topic messages.Based on the original SDG model,node fuzzy variables are added to solve the problems of low resolution and many false solutions of qualitative model fault diagnosis results.The publish/subscribe relationship between components is obtained from the middle layer of DDS,and the system interaction topology is restored,so as to solve the problem that the propagation path of system failure is covered due to the loose coupling of publish/subscribe communication model.(3)A fault diagnosis method based on fuzzy layered SDG model is proposed.Firstly,in order to improve the fault diagnosis efficiency of SDG model,model stratification method is used to stratify the model to reduce the fault search space.Then,based on the model's compatibility theory and fuzzy reasoning mechanism,the candidate fault sources of the system are inferred.Finally,the minimum-maximum synthesis method is used to calculate the fuzzy support degree of the fault,and the candidate fault sources are further corrected by comparing the threshold values to diagnose the root cause of the fault.(4)Based on the above proposed fault diagnosis scheme,a health management prototype system of distributed publish/subscribe system is designed and implemented.Finally,the fault diagnosis scheme of publish/subscribe distributed system is proved to be available by the function and performance tests of the prototype system.