Using Tadpoles to Reduce

A parallel variant of breadth-rst search for distributed computing is presented. The variant allows exhaustive enumeration of elements of a search space (implicitly deened graph) in which the representation of all graph nodes would otherwise require more than the total available memory. This algorithm requires the use of a tadpole data structure to partition the search space into connected subgraphs, with each subgraph stored within the memory of a single processor. Thus, graph the nodes are stored in a way that adds greater spatial locality, thereby reducing communication among processors. A parameter is deened that allows a linear tradeoo in which memory and communication are each reduced linearly as CPU time grows. The result appears to ll a gap in the literature, which has concentrated on parallel depth-rst search, but has been relatively sparse in the case of parallel breadth-rst search.