Non-Weighted Interface Specific Routing for Load-Balanced Fast Local Protection in IP Networks

Abstract

As a failure occurs, the affected traffic is quickly rerouted to backup paths for a network performing a fast protection scheme. Such a prompt reaction is aimed to reduce the damages caused by a failure. However, in some cases, the rerouted traffic may cause congestion along the backup paths which would lead to more packet losses than purely discarded affected flows. In this paper, we propose a load-balanced fast local protection scheme called Non-Weighted Interface Specific Routing (NISR) for determining the working and backup routing tables of IP routers. We jointly consider protection switching time, network survivability, and traffic load distribution together in the proposed scheme. In NISR, once a failure occurs, only the nodes adjacent to a failure divert affected traffic to backup paths. This local reaction process guarantees fast protection switching and reduces failure recovery time. Unlike the conventional IP routing, our approach relaxes the shortest path routing in computing working and backup routing tables. Most importantly, each interface in a router has its own routing tables. Combining the interface specific routing with the shortest path relaxation provides greater routing flexibility to enhance network survivability and load balancing. We formulate this as a mixed integer programming problem in which the traffic load on the most congested link is to be minimized. Since this problem is intractable by its NP-hard nature, we further decompose it into several sub-problems which are solved optimally and their solutions combined to provide a solution to the original problem. We perform experiments on some benchmark networks and compare the proposed scheme to several well-known schemes (including LFA, ECMP, OSPF, and NLB) on survivability ratio, link load distribution, and average path length for both normal and failure states. Through numerical results, we delineate that the proposed scheme achieves a sub-optimal solution, which is better for its high survivability and load balancing at the expense of slightly raising the average path hop count.