IP Fast Rerouting for Single and Multi-Link Failures

Abstract

The resilience of networks is increasingly becoming crucial as the adoption of cloud- based computing and interactive multimedia applications accelerates. Advances in optical technologies are enabling these modern network applications. The optical backbone, however, remains vulnerable to several types of failures that can cause considerable data loss. Natural phenomena, unintentional cutting of cables, configuration errors, or sabotage can cause multiple simultaneous link failures in IP networks. To reap the benefits of modern network applications, the network must provide resilience guarantees. We develop three novel IP network fast reroute methods in this dissertation. First, we address the problem of guaranteeing single-link failure recovery, while simultaneously load balancing traffic with negligible increase in path length. Existing failure recovery methods do not load balance traffic, or when they do they significantly lengthen paths. We turn our attention next to the problem of guaranteeing recovery from multiple link failures. The second IP fast reroute method we propose guarantees packet delivery under arbitrary link failures, provided that the number of link failures is one less than the size of the minimum edge-cut. None of the existing fast reroute recovery methods guarantee recovery from more than two-link failures. As with all fast reroute methods, we only use pre-computed and pre-installed routing table entries, and a fixed-length packet header to recover packets in the fast data plane. Our third IP fast reroute method provides a stronger failure recovery guarantee. We guarantee the delivery of a packet, provided that the number of failed links is at most one less than the number of link-disjoint paths connecting the packet's source to its destination, even when the failures disconnect the network. Comprehensive evaluations in real- world network topologies demonstrate our methods' excellent path length performance. Our methods may be implemented in modern networks, including SDNs, to provide failure recovery guarantees while retaining packet delivery speeds.