Automated learning of loop-free alternate paths for fast re-routing

Upon failure detection, IP networks need to reconfigure their routing and forwarding tables. Typically this task is executed by routing protocols such as Open Shortest Path First (OSPF). However during these short re-convergence periods, transient loops can occur, resulting in datagram loss of affected traffic flows. Several techniques have been developed to reduce the resulting datagram loss or to avoid this situation. However these approaches are not always able to cover all potential routing loops and can involve additional configuration. In this paper we suggest an alternative approach which relies on configuring loop-free alternate forwarding entries using learning techniques to reduce the configuration and setup effort, while still offering high speed switchovers upon failure events with minimal datagram loss resulting of transient loops. We show in a simulation environment that improved results can be obtained with respect to the number of link failures that can be covered, the resulting probability on having cycles in the alternate routing, the resulting quality of the alternate routing, and the induced communication cost of the learning procedure.