A fault recovery system that is fast and reliable is essential to today's networks, as it can be used to minimize the impact of the fault on the operation of the network and the services it provides. This paper proposes a methodology for performing automatic protection switching (APS) in optical networks with arbitrary mesh topologies in order to protect the network from fiber link failures. All fiber links interconnecting the optical switches are assumed to be bidirectional. In the scenario considered, the layout of the protection fibers and the setup of the protection switches is implemented in nonreal time, during the setup of the network. When a fiber link fails, the connections that use that link are automatically restored and their signals are routed to their original destination using the protection fibers and protection switches. The protection process proposed is fast, distributed, and autonomous. It restores the network in real time, without relying on a central manager or a centralized database. It is also independent of the topology and the connection state of the network at the time of the failure.

In this paper we show that even if every node or edge in an N-node butterfly network fails independently with some constant probability, p, it is still possible to identify a set of \Theta(N ) nodes between which packets can be routed in any permutation in O(log N ) steps, with high probability. Although the analysis is complicated, the routing algorithm itself is relatively simple. 1 Introduction This paper studies the ability of a network called a butterfly to route packets when many of its nodes and edges fail at random. A 32-node butterfly is shown in Figure 1. This network has been studied extensively and it, or one of its variants, has served as the routing network in several parallel computers. More information about the structural and algorithmic properties of butterflies can be found in the book by Leighton [13]. Some of the parallel computers that use butterfly networks are described in [7, 10, 18, 21, 24]. The algorithm described in this paper routes packets in a store-a...