Local-area network reliability models

Abstract

This paper describes analyses of reliabilities of local area networks (LANs) designed according to the most common IEEE 802 standards: the token ring architecture (IEEE 802.5, basically the IBM token ring architecture), carrier sense multiple access/collision detection (CSMA/CD) bus architecture (IEEE 802.3, basically the Ethernet architecture) and token bus architecture (IEEE 802.4, the recommended LAN architecture for the manufacturing atomation protocol, MAP). Each of these architectures is assumed to be implemented using electrical conductors; newer architectures using fibre optics (such as FDDI) are currently being studied. The primary reliability parameter studied is average source-destination availability. The models are based on information in the standards plus implementation details, using more implementation details than previous reliability models. Approximate analyses, which reflect all architecture details that appear to be pertinent to availability, have been developed, rather than exact analyses which ignore fine details. One conclusion is that details neglected in previous models have major effects on, and may be the most important factors determining, source–destination availability. At present, no measured values for the relevant parameters are obtainable, however, so all of our computations are based on arbitrary ‘guesstimates’ of these parameter values. A few other conclusions can be reached. Since major components of the bus networks (CSMA/CD and token bus) are similar, reliabilities of these networks are also similar, with differences determined primarily by differences in parameters. Adequate data to describe these differences were not obtainable, so our example computations for the two bus networks yield essentially the same results. Although its use of an active communications medium, instead of the passive bus used by the other architectures, may decrease reliability of the token ring, the token ring architecture gives more flexibility in bypassing or recovering from error situations. In some situations this could make the token ring the most reliable LAN, but ‘reasonable’ choices of parameter values could make any of the LANs most reliable. If many stations are inactive, however, any reliability advantage for token rings would be eliminated by even a small probability that inactive stations are not successfully bypassed