Internet failure protection using hamiltonian p-cycles found by ant-like agents

Abstract

'Link and node failures in the Internet are handled by restoration, i.e. by finding a new routing pattern. This may incur a considerable delay before the traffic flow may be reestablished. An immediate fault handling, i.e. protection, is desirable and required by some services and users. Protection cycles are a well known dependability measure and commonly applied In SDH networks with ring topologies. Applying protection cycles (known as p-cycles) in meshed network has recently been shown to provide good protection against failures. To be consistent with the Internet operational strategy, it is necessary to establish such rings dynamically, on line and In a distributed manner. Swarm intelligence is a distributed system construction concept which has shown promising results when applied to problems within the area of network and operational management. Swarm Intelligence systems are fully distributed (without central components), have high redundancy, are adaptable, and can find near optimal solutions to NP-hard problems. This paper describes a new swarm intelligence system based on the CE-ant algorithm developed by Helvik and Wittner. The system Is capable of finding near optimal Hamiltonian protection cycles in meshed networks. These are suited for protection in small to medium sized networks (ASs). To test and stress the concept, a case study is carried out mainly based on the topology, link loads and capacities of a real, diverse and rather large network. Results indicate that the agent system has a good ability to find good candidate p-cycles, and that in the network used as a case, approximately 80% of the link failures may be fully protected by the otherwise unused capacity in the network. With a good choice of p-cycle, even the very high capacity and high load links may have a substantial protection. Indexc Terms-Dependability, Protection in Internet, p-Cycles, Swarm Intelligence, CE-ants.