next up previous contents
Next: Overview of rerouting Up: MPLS_trees Previous: Contributions of this thesis   Contents


Resilience and protection in networks

Resilience refers to the ability of a network to keep services running despite a failure. Failure can have many causes, and we will focus in this thesis on ``fiber cuts'' or ``link cuts'', which result in the loss of all traffic that is forwarded on a failed link. Link cut is a common failure. For instance, in telephone networks, link cuts are the largest cause of service interruption time. Indeed, in a study conducted from April 1992 to March 1994, Kuhn [44] showed that cable cutting was the cause of 25% of telephone networks downtime.

Resilient networks recover from a failure by repairing themselves automatically. More specifically, failure recovery is achieved by rerouting traffic from the failed part of the network to another portion of the network. Rerouting is subject to several constraints. End-users want rerouting to be fast enough so that the interruption of service time due to a link failure is either unnoticeable or minimal. The new path taken by rerouted traffic can be either computed at the time failures occur or before failures. In the second case, rerouting is said to be pre-planned. Compared with recovery mechanisms that do not pre-plan rerouting, pre-planned rerouting mechanisms permit to decrease interruption of service times but may require additional hardware to provide redundancy in the network and consume valuable resources like computational cycles to compute backup paths. The different techniques we present in this chapter illustrate the trade-off between recovery speed and costs incurred by pre-planning.

In this chapter, we review existing techniques that improve network resilience. We first present an overview of rerouting and formalize the notion of total repair time for a network that reroutes traffic after a link failure in Section 2.1. In Section 2.2, we describe lower (physical and MAC) layer rerouting techniques. Low layer rerouting techniques rely solely on hardware and are therefore the fastest rerouting techniques available. However, they also require expensive hardware redundancy. We compare rerouting at lower layers with rerouting performed by the network layer without pre-planning in Section 2.3. Network layer rerouting is purely implemented in software and is therefore slower than lower layer rerouting. However, network layer rerouting does not use pre-planning, hence saving costs in hardware and CPU cycles compared with lower layer rerouting. Performing rerouting between lower layers and the network layer with ATM or MPLS presents a trade-off between recovery speed and pre-planning costs. In Section 2.4, we give an overview of MPLS Fast Reroute, a unicast rerouting technique that takes advantage of this trade-off. As shown in Section 1.3.1, taking the tree topology of multicast groups into account for routing purposes leads to cost savings in terms of computations and bandwidth. As a result, techniques specifically designed for multicast traffic have been developed. We describe these multicast rerouting techniques in Section 2.5.


Subsections
next up previous contents
Next: Overview of rerouting Up: MPLS_trees Previous: Contributions of this thesis   Contents
Yvan Pointurier 2002-08-11