Evidence combination for traffic adaptive routing

We propose a novel traffic adaptive routing algorithm to achieve equal path utilization in dynamic networks. We choose traffic measured at each router and the trip-time measured between networks as the indicators of the network condition. We derive the states of the equal-cost paths by combining the evidences pointed by the traffic and trip-time measures using Dempster-Shafer (DS) theory of evidence. The DS theory computes a belief measure and takes into account any ignorance present in the information sources acquiring the evidences and the inherent uncertainty in the network paths due to their dynamic nature. Based on the states of the equal-cost paths determined using DS theory, the routers route data packets among the equal-cost paths proportional to their load conditions. For example, if equal-cost paths P₁ and P₂ as seen from router R are in low and high load condition respectively, the router R route 2/3^rd of the data packets through P₁ and 1/3^rd through P₂. We modified the LINUX IP stack for adaptive routing and traffic measurements. The routing tables are updated every 30 seconds based on the states of the equal-cost paths determined using the DS theory. We demonstrate the algorithm in an interconnected network with LINUX machines configured as routers and show that the proposed evidence combination framework result in improved path utilization and compare the performance with the per-packet mode of traffic distribution.