Using probability network to infer link-level loss rate

It is important to obtain link-level performance data, such as loss rate and delay on each segment, which can help us to understand dynamic features of network traffics and identify bottlenecks. This knowledge also assists us to build and construct networks for better performance. Using end-to-end measurement to find link level performance has advantages in eliminating the burden from routers or switches. Instead of using classic statistical methods, such as maximum likelihood estimation, we use the probability network (a direct acyclic graph) to achieve the same goal. In addition, given a loss model, such as Bernoulli, the probability method can dynamically adjust the parameters to trace the traffic change. Simulations based on ns2 were conducted and the data received from the simulations were inferred by a corresponding probability network; the result is very encouraging.