Bottleneck bandwidth measurement based on kernel density estimation with variable window bandwidth

摘要

As Internet sustains high-speed development, its service quality, reliability and efficiency are taken into account once again. Some solutions are proposed. Network measurement technology can reflects operation condition of IP network and discover network abnormity in time which provides reference for orientation and discovery of network fault. Firstly, interval algorithm of packet pair and packet train in IP network bottleneck bandwidth is discussed and the impact of background traffic to bottleneck bandwidth is analyzed. A bottleneck bandwidth filtering algorithm with variable window width is proposed based on research of traditional measure sample filtering algorithm. This algorithm calculates window width of kernel estimation based on bandwidth range of initial detection and fluctuation value of measure samples. The more decentralizing the measure samples are, the larger the window width is which ensures calculated accuracy and reduces the estimated time.