Research On Available Bandwidth Measurement Method

With explosive expanding of Internet, the structure of Internet is becoming more and more complex. It is hard to access and acquire information about the network behavior. Network measurement is crucial to understanding the Internet network behavior and Internet performance. And it can use to design appropriate control mechanisms for better network performance. This thesis focuses on a fundamental measurement of network performance parameter: available bandwidth.Both in wired network and wireless network, an accurate, light load and short time estimate of the available bandwidth along a network path are useful in many applications. Available bandwidth estimation may help user to optimize route selection in overlay network, automatically adjust the transport bit-rate for video and audio stream in multimedia network application, access point adjustment in wireless network.Several measurement techniques and tools based on probe gap model (PGM) and the probe rate model (PRM) have been developed. The PGM is base on the assumptions that the tight link in the network path is the narrow link and the queue of bottleneck router is not empty between departure of the first probe and the arrival of the second probe. And the capacity of the link needs to be measured with capacity measurement tool firstly. PGM and PRM are both measure available bandwidth by send packets flow at rate higher than the available bandwidth, which may bring high load to network and impact normal network flow, especial low speed network and wireless network.This paper performed the experiments aimed at finding the relationship between probe packet flow and available bandwidth. We discover different phenomenon when send packets flow at rate lower and higher than the available bandwidth, and summarize an available bandwidth measurement approach which base on non-linear regression analysis to detect the congestion point base on the phenomenon. This available bandwidth measurement method overcomes the assumption restrictions of the PGM. At the same time, it overcomes the PRM problems such as high network load, low accuracy and iteration cycle is too long. Meanwhile, we performed performance comparison between this approach with other current available bandwidth measurement tools in the 10 Mbps and 100 Mbps laboratory network. In the comparison of accuracy and the average consumption of network, the available bandwidth Measurement base on non-linear regression analysis performed higher accuracy and relatively low consumption of the network.