An improved reliability approximation method is presented which reduces the number of states to be considered by exploiting the path structure of the network. The method does not impose restriction on the reliability criterion; criteria more general than those based on connectivity can be used. The reduction of the number of states is usually substantial, sometimes dramatic. The approach is applicable for all communication networks. Furthermore, very general reliability criteria can be used.>

A model is presented for the reliability and performance analysis of systems where components can go through degraded states in a statistically dependent manner. The model, called the cause-based multimode model (CBMM), is based on the idea that deviations of components from the 'up' state have underlying physical causes which can be explicitly identified and are statistically independent. The effects of several causes can be combined in a flexible manner. System reliability/performance measures can be computed by an approximation method, based on the consideration of the most probable states. Such states can be efficiently generated by using algorithms developed for an earlier multimode, independent-failure model by S.N. Chiou and V.O.K. Li (see IEEE J. Selected Areas in Commun., vol.SAC-4, no.7, p.1156-61 (1986)).>

Double-loop networks have a higher connectivity and therefore a higher potential for fault tolerance than single-loop networks. The authors present distributed routing schemes that fully utilize that fault-tolerance potential, and that are applicable in particular to a specific class of double-loop networks, characterized by a forward loop connecting all the adjacent nodes, and a backward loop connecting nodes separated by a distance that depends on N, the number of nodes. A nice feature of these schemes is that for a given node, only local knowledge on the status of the neighboring nodes and links is required. Yet the schemes detect faulty nodes and links and adapt to the situation, so that a packet will eventually reach its destination, if there exists a path. A simulation has shown that the average overhead resulting from the schemes, in terms of number of hops, does not exceed 17%, for values of N around 16.>