Proformance Analysis And Improvement Of Distributed Publish Subscribe System

With the rapid development of the Internet,the number of large-scale distributed applications increased rapidly,and the traditional mode of communication has been unable to meet the needs of the current large-scale,highly concurrent communications.Distributed publish-subscribe systems,which have many attractive characteristics including space and time decoupling,excellent scalability and availability and great adaptability for a distributed system with its distributed architecture,have received more and more attention.However,due to the complexity of communication between distributed systems' modules,the heterogeneity of users' softwares and hardwares,the differences between the time and the space of participants in the system and the scarcity of large-scale distributed research environment for distributed systems,the distributed publish-subscribe system performance research is always difficult.Now most researches on distributed publish-subscribe systems performance use benchmark.This method of getting the indicators of system performance has lag and is costly.In this paper,we use the way of building an abstract mathematical model to make a universal evaluation model to evaluate the performance of distributed publish-subscribe systems based on queuing theory.Firstly,we abstract the system step by step,building the network topology model,the business logic model and the mathematical model of the distributed publish-subscribe system sequentially.Secondly,we make some reasonable model assumptions to simplify the real environment to get a quantifiable queuing system.After that,we use a twice superimposed 2-state Markov modulated Poisson process to simulate the message input stream and get a MMPP/G/1 queuing model finally.After the validation of the model,we confirm that the model could forecast the performance of the model accurately when takes the internal and external factors which would affect the system performance into consideration and is also helpful to the discovery of the performance bottlenecks.The model not only offering a universal evaluation model of distributed publish-subscribe system,but also achieving a pre-deployment performance forecasting and providing guidance to improve system performance.