End-to-end available bandwidth estimation methodology

Abstract

A majority of current bandwidth estimation methodologies rely on the principle of the bottleneck spacing effect. Based on the concept of packet dispersion, many packet pair/packet train techniques were presented to estimate capacity/available bandwidth. However, these methods failed for measurement on high capacity path, because they could not measure bandwidth beyond the source node's maximum sending rate. In addition, current methodologies do not consider the effect of cross traffic routing on available bandwidth estimation. This paper analyzes the effect of the routing of cross traffic packets on available bandwidth measurement in detail. Then based on Monte Carlo Method, a novel methodology fundamentally different in the basic idea from the previous methods is presented to measure end-to-end available bandwidth. This method sends small single packet randomly instead of sending packet pair/train back-to-back. It could work on network whose capacity is far beyond the maximum sending rate of the sender. Analysis and simulations show that besides end-to-end available bandwidth, this method could measure the capacity and idle ratio of targeted link, and then calculate the change of traffic flow on each node and the percentage of different cross traffic on each link.