Delay-based Measurement And Analysis Of Bottleneck Links In The Internet

With the rapid development of the Internet, a variety of network applications haveemerged and people’s life is heavily depended on the Internet. Network performancehas become one important problem for the Internet Service Providers (ISPs) andresearchers. Bottleneck links represent the handicaps that prevent the improvement ofnetwork performance. It has become an important research topic on how to detect andeliminate them. Currently, the most widely-used methods use available bandwidth in themeasurement and analysis of bottleneck links. But, there are few approaches onmeasurement and analysis of bottleneck links using network delays.This thesis focuses on the study of bottleneck links determined by network delaysin the following aspects.Firstly, the thesis introduces and analyzes the related research work in detail.Existing methods have not considered multiple bottleneck links in an end-to-end path orhave common problems of poor generality. This thesis proposes two new approaches toidendify multiple bottleneck links using data clustering methods. One is based on Fisherclustering method, and the other one is based on Ward clustering method.Secondly, in order to verify and compare the generality and accuracy of themethods, and use them to analyze the distribution characteristics of bottleneck links,two groups of experiments are conducted as follows. In the first experiment, wemeasure10different destinations in the Internet for14days using an improvedtraceroute program which probes many destinations in parallel based on OpenMP. Theexperimental results have shown that the proposed methods can accurately locatemultiple delay bottleneck links in most cases. In the second experiment, we measure77different destinations from30probing sites on Planetlab for30days, the results are usedto analyze the distribution characteristics of bottleneck links in the Internet.At last, using the proposed approaches, the thesis has found that bottleneck linksare mainly few fixed links which are in the intermediate of end-to-end paths or near thedestinations. Normal changes of routes will make little difference to the bottleneck links.Furthermore, the results have shown that bottleneck links often occur as inter-AS(anonymous system) links or close to inter-AS links. The special deployment strategy ofBGP routers in an AS may lead to the results. Besides, the results have alsodemonstrated that bottleneck links may often appear as inter-continental links orsatellite links, and physical lengths of links are chief factors for delay bottleneck links inend-to-end paths.