The Research Of Bandwidth Measurement Techniques In IP Network

With the rapid development of Internet, its quality of service, reliability and efficiency are taken into account over again, and lots of solutions are brought forward. One of them is network measurement, which can give many quantitative metrics reflecting operating status of IP network, revealing upgrading or degrading of network performance, and discovering network outlier in time. The network bandwidth, as one of the most crucial network resources, directly influences user-perceived performance, and its accurate measurement and estimation is an essential problem in traffic engineering, as well as other metrics such as delay and packet loss.In network layer link bandwidth and available bandwidth capture the relevant properties of network path, which is baseline of performance management and QoS control. This paper concludes the present models and approaches of bandwidth measurement and compares performance and applicability between them. Furthermore, an improved approach to measure the bandwidth of specified link and an algorithm to locate the path bottleneck are presented. And available bandwidth provides a dynamic measure of the load on a path that makes its estimation difficult. This paper implements the estimation of available bandwidth and gives its bound through delay analysis to measurement stream based on the Cruz traffic model. Also the problems and solutions in the measurement implementation are analyzed.However, users often focus more on the real performance of application in practical requirements. Good bandwidth guarantee in network layer does not always means good application performance. This paper defines transact bandwidth to refer to service throughput in application layer. By emulating the operating process of clients accesses to server, the transact throughput is obtained through the request-reply mode and the capacity of end-to-end path to support relevant service is acquired. By collecting experimental data, we analyze the composition and distribution of HTTP service response time; therefore investigate the high variation of the download rate in multiple accesses to the same web page. It proves that some popular performance fault detection method, such as mean-standard deviation calculation, is not fully applicable to analyze this case. Then a boxplots-based algorithm to detect HTTP metric outliers is presented to monitor the performance of HTTP service bandwidth.