| With the explosive growth of Internet, performance becomes a big issue for Internet content providers. Popular World Wide Web sites receive tremendous requests from the clients all over the world, which causes heavy Internet traffic and long response time. Placing mirror servers is a solution to improve the performance. In this study, I investigate the mirror placement problem on the Internet. Firstly, Self-Organizing Features Maps (SOM) clustering, K-means clustering and Out-Degree analysis methods are employed to identify mirror server candidates. Secondly, the Ant clustering algorithm and the 0-Greedy algorithm are employed to choose the mirror servers. Finally, the performance of the mirror candidate selection algorithms and the mirror placement algorithms are evaluated by using RTT and hop count as the distance metrics. The results show that K-means and Out-degree give very similar and good performance results as mirror server candidate selection algorithms. Moreover, 0-Greedy algorithm gives better results as a mirror placement algorithm than the Ant clustering algorithm. |
