External memory breadth-first search with delayed duplicate detection on the GPU

  • Authors:
  • Stefan Edelkamp;Damian Sulewski

  • Affiliations:
  • TZI, Universität Bremen, Germany;TZI, Universität Bremen, Germany

  • Venue:
  • MoChArt'10 Proceedings of the 6th international conference on Model checking and artificial intelligence
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

We accelerate breadth-first search by delegating complex operations to the graphics processing unit (GPU). The algorithm exploits external memory: if the state space becomes too large to be kept in main memory, it is maintained I/O-efficiently on disk. As in many other approaches for external memory graph search, we apply delayed duplicate detection. The search proceeds in breadth-first layers with increasing minimum distance from the start state. For each layer stored on disk, we load chunks into the systems memory, which are forwarded to the memory on the graphics card. Here we test if outgoing transitions are enabled and generate all successors. Finally, we eliminate duplicates delayed by sorting on the GPU. Even facing the overhead of I/O access, noticeable overall speed-ups are obtained.