A New Method for End-to-End Available Bandwidth Estimation

Previous Probe Gap Model (PGM) based available bandwidth (AB) estimation methods all request the "busy assumption" that probing packet pairs should be in the same busy period when transmitted on bottleneck link, which is hard to satisfy especially for the low utilization path. In this paper, we first present a new probabilistic methodology to estimate AB under "non busy assumption". The methodology is quite accurate on the low utilization network path. Secondly, we propose a metric to weigh the busyness of a network path based on the distribution of output probe gap. Using the metric, we combine our new methodology and previous methodology, and present a new AB estimation method called Adaptive Available Bandwidth Estimation (A_ABE) which is fit for both low utilization and high utilization paths. We use NS-2 simulation and reproduce traffic from real Internet links to evaluate A_ABE. Compared with previous methods, A_ABE shows its advantages in terms of accuracy, overhead, and also the robustness when confronted with non-persistent cross traffic in multiple hop situations.