The Analysis Of Internet Packet Delay Behavior And Active Measurement Techniques

The Internet has been the information infrastructure of the human society after about 30 years development; however people only have a superficial knowledge about the Internet yet. The expanding of the Internet along with the sharp increasing of the users and the emergence of various kinds of network applications put more and more serious challenges in front of network monitoring and management. Network measurement is a basic and important means for network monitoring and management. Further more network measurements will be significantly helpful for deeply understanding of network behavior, reasonably dimensioning of network resource and better designing of network protocols.The dissertation focuses on three issues in the area of measurement and analysis of the Internet, namely the Internet packet round-trip time (RTT) dynamics, active measurement sampling techniques, automatic cooperation measurement technique and measurement system. The major contributions of this dissertation are summarized as follows:(1) The dissertation studies the Internet packet round-trip time dynamics in small time scale, based on high density network measurements with interval of 10ms. And it is found that the Internet RTT series exhibit two completely different characteristics, namely spiky and normal characteristics, the spiky part of the RTT series can be characterized by sharply increasing and linearly decreasing. A phase plot based separation algorithm is proposed to eliminate the spiky elements from the original series. And then we investigate the impact of the spiky elements on the scaling behaviors of the original series by using the separation algorithm and DFA method (Detrended Fluctuation Analysis) and draw the following conclusions: the original Internet RTT seires can not be characterized by a simplex scaling behavior; although spiky elements account for a very little propotion of the original RTT series, they do influence the scaling behavior of the original RTT series, and when spiky data are eliminated, the normal elements of Internet RTT series can be characterized by Long-Range Denpendence with Hurst exponent from 0.55 to 0.8. And the above conclusions are validated by numeric experiments.(2) The dissertation systematically analyze and evaluate the performance of active measurement sampling techniques based on theory analysis, measurement data analysis and simulation experiments. And the dissertation focuses on estimating the packet delay performance experienced by user packets. We first propose an active measurement sampling model and then analyze the model by using discrete ASTA (Arrivals See Time Averages) theory and draw the following conclusions: the estimation will be unbiased when and only when active measurement procedure is not correlated with the measured packet delay process;...