DFA minimization in map-reduce

Abstract

We describe Map-Reduce implementations of two of the most prominent DFA minimization methods, namely Moore's and Hopcroft's algorithms. Our analysis shows that the one based on Hopcroft's algorithm is more efficient, both in terms of running time and communication cost. This is validated by our extensive experiments on various types of DFA's, with up to 217 states. It also turns out that both algorithms are sensitive to skewed input, the Hopcroft's algorithm being intrinsically so.