Iterative mapreduce based heuristic algorithm for solving N Puzzle

Abstract

Mapreduce Programming paradigm provides an elegant and efficacious parallel implementation of Heuristic Search Algorithm. We present here a Parallel Breadth First Heuristic Search (PBFHS) Algorithm for solving very large combinatorial problems. Using N Puzzle as our application domain we found that scalability of Breadth First Search (BFS) and Iterative Deepening A* (IDA*) is limited on a single machine due to hardware constraints. We generated a highly restrictive, large search space using combination of three efficient and admissible heuristics namely Manhattan Distance, N-Max Swaps and Tile out of row and column. A Hadoop cluster comprising of 12 processors, solves the hardest 15 Puzzle in 2 hours, and 24 Puzzle in 6 hours of computing time.