On the pathological search behavior of distributed greedy best-first search

Abstract

Although A∗ search can be efficiently parallelized using methods such as Hash-Distributed A∗ (HDA∗), distributed parallelization of Greedy Best First Search (GBFS), a suboptimal search which often finds solutions much faster than A∗, has received little attention. We show that surprisingly, HDGBFS, an adaptation of HDA∗ to GBFS, often performs significantly worse than sequential GBFS. We analyze and explain this performance degradation, and propose a novel method for distributed parallelization of GBFS, which significantly outperforms HDGBFS.