Internet Active Bandwidth Measurement Study

The 21st century is an internet-based information age. With the growing applications of internet, people tend to set higher standards for its performance while enjoying the efficiency and convenience coming with it. Bandwidth is one of the most important internet performance indexes. Therefore, to measure bandwidth dynamically, accurately and quickly is of great importance in improving internet performance and its utilization ratio. Based on a wide range of relevant theories and principles, till now, quite a few measurement tools have been developed and put into use. However, due to the modeling simplification or the constraints of measurement methods, and the limitations of measurement hardware, these tools have such weaknesses as unsatisfying accuracy, complex procedures, limited application, etc.Based on a large body of literatures home and abroad, this thesis aims to conduct a research on bandwidth measurement. Firstly, it reviews the fundamental concepts concerning bandwidth measurement, as well as relevant methods and models, mainly VPS model and PGM model. Then it identifies the current problems existing in bandwidth measurement. Finally it proposes solutions to the problems in asymmetric links and equipment deployment, whose feasibility is then tested by theoretical analysis and NS2 simulation experiments. The main contributions of this thesis are as follows:Firstly, the thesis presents an approach to measure the two-way bandwidth of asymmetric links. It is a single-ended measurement algorithm, originating from VPS models. In this approach, the source of the measured path sends a tetrad probe packet composed of TCP and ICMP packets, which then produces a response packet after reaching the destination end. RTT latency of each probe packet is calculated after the source receives the response packet. Then, the upstream and downstream bandwidth of the link is calculated by analyzing the composition of RTT latencies and examining the linear relationship between the differences in packet size and those in RTT latency.Both theoretical analysis and simulation experiment are employed to test the new approach, which finds that it can reduce the probe flows injected into the network, accelerate measurement process, and improve measurement accuracy, thus solving the problems of asymmetric links in real-world internet paths.Secondly, in order to overcome the difficulties in the deployment of measurement equipments, the thesis presents a PGM-based approach to measure the bottleneck bandwidth of sub-paths. As a double-ended measurement algorithm, this approach can measure the bottleneck bandwidth of a sub-path at the two ends of a hyper-path including this sub-path, and meanwhile locate the bottleneck link, by using packet trains composed of probe packets and TTL limited load packets, where the probe packets are at the two ends of load packets. The well-designed probe packets are sent from one end of the hyper-path, and in the process, load packets are lost due to the limited TTL on the sub-path. In this way, the time spacing between the two probe packets can be preserved to the destination end. There the time spacings between probe packets are calculated and the largest one is corresponding to the time spacing of the bottleneck link on the sub-path. Thus, the bottleneck bandwidth can be calculated by time spacing and packet size. The bottleneck link can be located as well. Finally, this approach is tested with NS2 simulation tools in different cross traffic, which concludes that this approach is both effective and feasible.