Preprocessing Parallelization for the ALT-Algorithm

摘要

In this paper, we improve the preprocessing phase of the ALT algorithm through parallelization. ALT is a preprocessing-based, goal-directed speedup technique that uses A* (A star), Landmarks and Triangle inequality which allows fast computations of shortest paths (SP) in large-scale networks. Although faster techniques such as arc-flags, SHARC, Contraction Hierarchies and Highway Hierarchies already exist, ALT is usually combined with these faster algorithms to take advantage of its goal-directed search to further reduce the SP search computation time and its search space. However, ALT relies on landmarks and optimally choosing these landmarks is NP-hard, hence, no effective solution exists. Since landmark selection relies on constructive heuristics and the current SP search speed-up is inversely proportional to landmark generation time, we propose a parallelization technique which reduces the landmark generation time significantly while increasing its effectiveness.