Self-similar cross-traffic analysis as a foundation for choosing among active available bandwidth measurement strategies

Among the many active, available bandwidth estimation tools and methods that have appeared recently in the literature, none has emerged that is accurate and robust enough to function in different networking environments. One of the main reasons for this problem is the highly variable statistical nature of network traffic. We present a novel theoretical method for active measurement strategy analysis through self-similar process sampling analysis. The results obtained show a relationship between the main parameters in the measurement procedure, such as the number of samples, sample length, and sample distance. The study suggests that accurate measurements require the use of many samples and thus long measurement procedures, sometimes on the order of several minutes, depending on the amount of cross traffic. Additionally, through the analysis of several existing Internet packet traces, we recommend that the minimum single sample probe stream length must be longer than the average cross traffic interarrival. Following these findings, we propose a new method for available bandwidth estimation that has shown significant accuracy under different network setups.