Chain reactions in networks

A set of situations that are susceptible to network disasters is identified. Several examples of actual breakdowns are presented. Although they occurred in different types of networks, they have grant similarities. Several examples of potential breakdowns that can result from very subtle design failures are also given. The discussion supports the belief that distributed programs in general can have very subtle design failures that are hard to detect. However, it is shown that although it is theoretically impossible to prevent the berserk loops of messages that can bring down an entire computer network, a carefully applied set of rules will minimize these disasters.<>