External A* External A*

In this paper we study External A*, a variant of the internal A* algorithm that employs external memory. The approach applies to implicit, undirected, unweighted state space problem graphs with consistent estimates. The complexity of the External algorithm is almost linear in external sorting time and accumulates to O(sort(|E|) + scan(|V |)) I/Os, where V and E are the set of nodes and edges in the explored portion of the state space graph. Given that delayed duplicate elimination has to be performed, the established bound is I/O optimal. In difference to the internal design in the construction we exploit memory locality to allow block rather than random access. The algorithmic design refers to external shortest path search in explicit graphs and extends the strategy of delayed duplicate detection recently suggested for breadth-first search to best-first search. We conduct experiments with sliding-tile puzzle instances.