Dynamic Server Push Method Decision Technology Research Based On Machine Learning Algorithm

The traditional B/S architecture is based on the HTTP protocol and is a stateless communication method that gradually fails to meet the quality requirements of real-time communication.Server push technology is based on Ajax and Comet technology,enabling real-time communication between the web server and the client.According to the characteristics of different push modes,different application scenarios select different push modes.In this thesis,the analytic hierarchy process and the entropy weight method are both chosen to comprehensively evaluate the weight of environmental factors while the XGBoost algorithm based on CART decision tree is used to construct a dynamic decision model and an experimental scenario of online games is selected to train the decision module and perform performance test.The experimental results show that the XGBoost algorithm is applied to the dynamic decision-making problem of server push mode,which can basically meet the reasonable decision-making of push mode in the corresponding scenario,and can show good performance.It can utilize network resources reasonably compared with single push mode.The main contents of this thesis are listed as follows:(1)Explain the development status of the current server active push technology and the problems to be solved.The three different push methods have different usage scenarios.The polling is suitable for simple service scenarios with fixed request periods and low requirements for implementation.Long polling is suitable for general events with frequent event frequency and real-time requirements.The performance is moderate.Comet is of moderate performance,which is suitable for services with certain requirements for event frequency and real-time performance.Websocket is a new standard.It can also maintain long-term connection between server and browsing.However,it requires high background development and requires a separate background service program.(2)Analysing current artificial intelligence and machine learning algorithm.Choosing the XGBoost based on CART tree series algorithm as the decision algorithm for this thesis,XGBoost based on tree model can handle table data well,and also has some features that are not available in deep neural networks.(3)Using the XGBoost algorithm based on CART tree to design a model to realize the dynamic push server push function,the program realizes and tests the rationality of the decision and other aspects of performance.(4)Design different experimental scenarios to test the model designed in this thesis,and analyze and summarize the experimental results.