Modeling And Analyzing Of Unstructured Overlay Networks For Service Discovery Application

The rapid development of the Internet has promoted the trend of networked software. Network services have gradually become the core and the form of software. Web Services are important components of networked software, and the goal of service discovery is to search Web Services accurately and efficiently on the Internet. Due to the problems in centralized service discovery system such as single-point failure and poor scalability, distributed service discovery system based on unstructured overlay networks attracted extensive attention. Previous studies mainly focus on the system architecture and the description of Web Service. Unstructured overlay networks, the foundation of the system, have not been carefully studied. The thesis studies unstructured overlay networks for service discovery application by analyzing real topology, modeling networks and optimizing performance of searching.Firstly, it is studied what topological characteristics the unstructured overlay networks have, and then the network model is proposed based on the characteristics analysis. Gnutella is a typical unstructured overlay network. By analyzing real Gnutella topology, the heavy tailed degree distribution, the peak of degree distribution and the small-world characteristic of unstructured overlay networks are verified, and its weak Rich-club characteristic is discovered. The GFCC model for unstructured overlay networks is proposed by analyzing the internal mechanism of the characteristics.Secondly, the thesis proves the effectiveness of the GFCC model and reveals the relationship between nodes'behavior and topological structure in unstructured overlay networks for service discovery application. The fitness function is selected according to the characteristics of service discovery application. Then the degree distribution, the rich-club coefficients, the clustering coefficients and the average path length of the GFCC model are analyzed by theoretical derivation and simulation. Analysis results show that the GFCC model can recreate the topological characteristics of real networks. The results also reveal the influence of nodes'caching behavior, the number limit of neighbors and nodes'clustering behavior on topology structure.Finally, the thesis analyzes and optimizes the performance of Web Service searching using the GFCC model, and SOONET, a distributed service discovery system, is designed by applying research conclusions to practice. By analyzing and comparing the searching success rate , the searching delay and the searching cost of flooding algorithm and random walks algorithm in the GFCC model, k-Walkers is proposed as the suggested searching algorithm for distributed service discovery system. And the optimum copies number distribution of Web Service description is obtained by constrained optimization. Designed on the basis of the previous research conclusions, SOONET, as a Web-Service-based distributed service discovery system, implements k-Walkers algorithm by using token-based routing strategy of query message and optimizes the copies number distribution of web service description by caching Web Service description on search path.