End-to-end Available Bandwidth Measurement Methods

With the rapid development of Internet today, in order to keep the network running better, Internet Service Providers use a variety of network measurement tools to measure network performance criterions such as latency, bandwidth, packet loss rate and topology. Many researchers and organizations also committed to the research of the available bandwidth estimation. They have developed a lot of available bandwidth measurement tools. In this way they can understand the network internal structure and the operating conditions, so as to provide excellent management services and quality of service. In the network measurement area, the result of end-to-end available bandwidth estimation along a network path has very important practical significance and has become a hot issue in the Internet measurement area as it can be applied to dynamic server selection, admission control, overlay construction of peer to peer system, etc.The paper conducts an investigation in the available bandwidth estimation, and describes basic concept, measurement methods and measurement models of available bandwidth measurement, introducing the measuring principle and the advantages and disadvantages of several typical available bandwidth measurement tools such as pathload, pathchirp and abget systematically and comprehensively. According to the principle that we should make the estimation fast and accurate, and two typical available bandwidth measurement tools abget and pathchirp have been improved on the premise that we should minimize the intrusiveness of the probing packets on the cross-traffic:First, measurement tool for single-end host system abget which only requires access at one end of the measured path has been improved. Abget is based on TCP protocol, so it has a great advantage to perform measurements even in environments where ICMP and UDP packets are blocked by firewalls or rate-limited. Compared with double-end host system, although abget has better practicality, the convergence is slow and the measurement accuracy is not high. Against these shortcomings, this paper introduces the measurement methods which are fast and accurate in double-end host system to abget, improving the method of adjusting the rate of the probing packet train and the method of detecting the OWD trend. Simulation results show that the improved method is faster and more accurate than the original algorithm.And then, on the basis of the analysis of measurement tool for double-end host system pathchirp which requires access at both end of the measured path, indicating that pathchirp is superior with short measurement time and less intrusiveness to the cross-traffic, but the probing packets are easy to be interfered by the cross-traffic in the following links, so the measurement accuracy is not high. Against this shortcoming, this paper improves the structure of the chirp train. The improved probing packet train uses the variable region of the available bandwidth. In the region, packet instantaneous-rate sampling is more frequent and the number of packets at each sampling rate is larger. The tool uses the observed value as the center value of the region in the next estimation. So the improved probing packet train covers a range and contacts the several rounds of measurements for the path at a certain period of time with not so many probing packets. Simulation results show that the improved method is more efficient and accurate than the original algorithm with not adding load to the network path significantly.