Abstract
In IP networks, a series of round trip times is often fed to a minimum filter, which outputs the minimum among n recent sample values. This technique is indispensable in applications such as time synchronization and bandwidth measurement. The performance of a minimum filter is estimated by the difference between its output value and the true minimum of the round trip time; the performance is better if the difference is smaller. The performance of a minimum filter depends on the number of samples, among which the minimum value is selected. However, it is difficult to decide the optimal number of the samples beforehand because it depends on the conditions such as background traffic and measurement interval. This problem is solved by the adaptive minimum filter that changes the number of samples depending on the variation of measured values. The purpose of this paper is to evaluate the adaptive minimum filter through experiment. As a result, it is shown that the adaptive minimum filter performs well independently of background traffic and measurement interval.
