Finding the shortest path between any two nodes in a graph, known as the All-Pairs Shortest Paths (APSP), is a fundamental problem in many data analysis problems, such as supply chains in logistics, routing protocols ...
详细信息
Finding the shortest path between any two nodes in a graph, known as the All-Pairs Shortest Paths (APSP), is a fundamental problem in many data analysis problems, such as supply chains in logistics, routing protocols in IoT networks that involve consumer electronics as well as data analysis for social networking apps and Google Maps apps used by the general public on their smartphones. In this work, we present a novel approach to solve the APSP problem on multicore and GPU systems. In our approach, a graph is first pre-processed by partitioning the graph into sub-graphs. Then, each sub-graph is processed in parallel using any existing shortest path algorithm such as the Floyd-Warshall algorithm or Dijkstra's algorithm. Finally, the distance results in individual sub-graphs are aggregated to obtain the distances of APSP for the entire graph. OpenMP and CUDA are used to implement the parallelization on multicore CPUs and GPUs, respectively. We conduct the extensive experiments with both synthetic and real-world graphs on the JADE (Joint Academic Data Science Endeavour) cluster at the University of Oxford, which is part of the Tier-2 high performance computing facilities in the U.K. In the experiments, we compared our methods with three existing APSP algorithms in the literature, including n-Dijkstra, ParAPSP and SuperFW. The results show that our methods outperform the existing algorithms, achieving the speedup of up to 8.3x over Dijkstra.
暂无评论