Fast reroute paths algorithms

Abstract

In order to keep services running despite link or node failure in MPLS networks, RSVP-TE fast reroute (FRR) schemes use precomputed backup label-switched path tunnels for local repair of LSP tunnels. In the event of failure, the redirection of traffic occurs onto backup LSP tunnels that have the same quality of service constraints as original paths. Local repair of LSP tunnels notably differ from traditional (1:1) dedicated path protection schemes in that traffic is diverted near the point of failure which speeds up the protection process by not having to notify the source and then resend the lost traffic. This gain in protection delay is crucial for MPLS networks which would otherwise suffer from an important recovery latency.

In this paper, we investigate the algorithmic aspects of computing original paths along with their back-up so that they satisfy quality-of-service constraints (namely, delay) for single link or multiple link failure. In the case of single link failure, we propose an algorithm in O(nm+n ²log(n)) that computes shortest guaranteed paths with their backup towards a single destination. In the case of directed graphs, we show that this algorithm is optimal by proving that computing shortest guaranteed paths is as hard as to compute multiple source shortest paths in directed graphs. In the case of undirected graphs, we propose a faster algorithm with time complexity O(mlog(n)+n ²). We also provide a distributed algorithm based on Bellman-Ford distance computation which converges in 3n rounds at worst.