Cooperative Distributed Algorithms for Dynamic Cycle Prevention

Parallel distributed algorithms are presented for adding and deleting edges in a directed graph without creating a cycle. Such algorithms are useful for a variety of problems in distributed systems such as preventing deadlock or ordering priorities. The algorithms operate in a realistic asynchronous computer network environment in which there are numerous possible interactions among overlapping instances of the algorithms. The distributed algorithms are derived from a sequential algorithm. In developing distributed versions of the algorithm from a sequential version, the vital role of an invariant is emphasized. Global correctness of the distributed algorithms relies on (locally) preserving this invariant. Interactions and cooperation between various activations of the algorithms are exploited in order to minimize redundant computation.