Research Of Accurate Available Bandwidth Probing And Traffic Monitoring Technology Based On End-to-end Measurement

Efficient and reliable Internet measurement remains an important goal for many applications. For example, it is essential for audio and video streaming applications to know the available bandwidth of the network path. Network measurement is generally considered difficult to measure because of its dynamics, especially in the Internet environment. In this thesis, we surveyed the state of the art in end-to-end Internet measurements, reviewed the classifications and evaluation criterions of measurement methods.The packet-pair technique is the most popular end-based approach for bandwidth measurement. Most packet-pair based proposals use the fluid cross-traffic model to justify the design of their estimation techniques. However, real traffic is certainly not fluid. The consistency of the statistical packet-pair dispersion model in general bursty cross-traffic conditions has still not been fully justified in published research. To achieve high accuracy and efficiency in available bandwidth measurement, we contribute analytical insights into packet-pair dispersion techniques by developing a queuing model to describe the impact of the cross-traffic on the packet-pair dispersion under bursty cross-traffic conditions. We use this one-dispersion model to help understand the interaction between the probing packets and the competing traffic. We next present a rigorous formulation by using statistical terms to demonstrate the consistency of the packet-pair probing technique under conditions of non-fluid traffic, extending previous rationales that are based on fluid cross-traffic models.Furthermore, we note that available bandwidth measurements are also subject to"elastic bias", due to the effectiveness of traffic burstiness. The measurement variance determines the dynamics of the"elastic bias", but existing measurement techniques do not take account of the actual variance, which means that they are often subject to significant fluctuations in accuracy. Based on above queuing model, we track the boundaries of the measurement variance on different traffic distributions. We find an important result, which estimates the measurement variance irrespective of traffic types. All formulas are validated with ns simulations. Then using large-scale Internet measurement experiments, we present an explicit solution for measurement variance prediction on the Internet.We build an accurate and efficient available bandwidth measurement algorithm using adaptive packet-pair sampling to infer internal network status from end-to-end measurements. We show that the algorithm works correctly on the Internet and offers much more dependable measurement results than existing bandwidth estimation tools.Scaling behaviors of traffic influence network modeling, service providing and traffic engineering. In order to understand the causes of small-time scaling phenomena, an extensive wavelet analysis is performed to Ethernet IP Traces from 8μs time scale. Our results contain two interesting factors: Uncorrelated scaling is obvious on time scales lower than certain point which is referred as the"a point". a point is the starting point for correlated scaling behaviors and the position of a point can be calculated by bandwidth and MTU.Most Internet traffic measurement systems today use router-based passive monitoring. Passive measurement has several disadvantages. In this paper, we present a new technique to analyze and validate traffic characterization with packet train probing. We first construct a simple model to understand how competing traffic changes the probing packet gap in a train for a single-hop network. Based on this insight, we present a new theoretical model of traffic characterization based on a packet-train probing. We use extensive Internet measurements to show that in addition to provide robust estimators of the Hurst parameter, it can also do multi-fractal analysis with good precision. We finish the paper by showing, both analytically and experimentally, that the packet-train probing based proposals works correctly on the Internet and offers dependable measurement results.