Stability of Distributed Algorithms in the Face of Incessant Faults

For large distributed systems built from inexpensive components, one expects to see incessant failures. This paper proposes two models for such faults and analyzes two well-known self-stabilizing algorithms under these fault models. For a small number of processes, the properties of interest are verified automatically using probabilistic model-checking tools. For a large number of processes, these properties are characterized using asymptotic bounds from a direct Markov chain analysis and approximated by numerical simulations.