Research On Publish/Subscribe Distributed System Health Evaluation Technology

The publish-subscribe distributed system is widely used in the military and industry.At present,there are some monitoring software in the operation and maintenance management of the publish-subscribe distributed system,but they can only provide a large number of system operating parameters,and cannot visually display the health state of the system.These software require the operation and maintenance personnel to analyze the data according to experience in the later stage and evaluate the overall operational status of the system.Based on the characteristics of the publish-subscribe distributed system,the software health management technology is introduced into the publish-subscribe distributed system,the health evaluation and measurement methods of the publishsubscribe system is systematically studied,and the health assessment is divided into two parts: component health assessment and system health assessment.A stratified health assessment architecture of component and system is formed.In this thesis,a health evaluation model and method of publish-subscribe distributed system is proposed,by collecting system runtime data and calculating with Spark big data processing technology,the health status evaluation of the publishsubscribe distributed system is realized.The main research contents of the thesis include:1.Aiming at the problems of multiple states in the software running process,compared with the method based on distance and density,a component health measure method based on local outliers is proposed.The spatial distribution of software running data in various states is analyzed,and the density relationship of the data points to be measured relative to the historical data is calculated,thereby obtaining the degree of local deviation of the points to be measured,and the local outliers value is used to evaluate the health of the publish/subscribe component.2.Aiming at the characteristics that the exceptions of the publish/subscribe distributed system are transmitted with the chain of the publish/subscribe topic data,a system health evaluation method based on the topology relationship of component publishing and subscription is proposed.Since each publishing component has a connection similar to a webpage,and the number of topics published between the components can indicate the strength of the connection.In the thesis,the Page Rank algorithm is generalized to the component weight calculation,and the component weight is determined by the topology based on the publishing and subscription relationship.Component weights are then used to calculate the health of the publish/subscribe distributed system in a weighted manner.3.Based on the above-mentioned publish/subscribe distributed system health evaluation method,the Spark big data processing technology is used to design and implement a prototype system for publish/subscribe distributed system health assessment software.The functional modules of the system includes data collection,component and system health evaluation and web page display.The functional and performance test results of the prototype system show that the proposed method can effectively evaluate the health status of the publish and subscribe distributed system,and correctly reflect the data content and component performance anomaly in publishsubscribe distributed system.