Research And Realization Of Parameter Optimization Method In Web System Based On Genetic Algorithm

With more and more commercial network sites emerging and rapid increacement in user visits, Web servers require higher performance, and how to efficiently optimize the performance of Web systems under the conditions of limited resources has become a major concern in system development and maintenance. The performance of Web systems can be greatly improved by properly tuning their configuration parameters. However, manually tuning tens of configuration parameters and getting the right one in the complex Web deployment enviroment are an arduos and time-consuming task.In this thesis, we apply our method combined with genetic algorithm to optimize the configuration of a multi-layer Web application architecture, and find that it can quickly selects one of nearly best configurations with high probability. The main contents of this thesis are listed as follows:(1) The design and implementation of genetic algorithm. We analyse the characteristics of the existing optimization techniques, the solution of optimized problems and the factors that affect the algorithm performance. We in-depth study the basic principle and workflow of genetic algorithm, apply it to the Web system parameter optimization problems, encode the main tunable configuration parameters in Web server Tomcat, database server MySQL and Java Virtual Machine, generate intial population combined with Latin Hypercube Sampling approach, calculate the fitness value, do the operation including select, crossover and mutation, search for optimal solution iteratively, finally give the experimental results and the algorithm performance analysis.(2)The structure of experimental platform. We deploy the benchmark of TPC-W, construct Web application environment, and do works including the construction of online book sales website and the simulator which generates user requests, the coding of shell script language batch process which controls parameters modification, software reset, and log performance data to file, optimize overall system performance.(3)Genetic algorithm and QOG algorithm are combined to reduce the total measurement time of parameters optimization in Web system. Genetic algorithm can quickly approach to the optimal value and is more accurate and efficient compared to that QOG uses regression function to predict configuration parameters corresponding to the system performance. We combine genetic algorithm with QOG, and do some changes on QOG to reduce the total measurement time of parameter optimization.Tuning performance of Web system means to find the correct parameter settings under the specified conditions. We verify the practicality and efficiency of our optimization algorithm by running performance test and program in Web application environment. All of those prove to be practical value and social benefits.