Accelerating BFS shortest paths calculations using CUDA for Internet topology measurements

Within the last decade, the number of devices connected to the Internet has seen immense growth and it has grown to be a large and complex network. To analyze this network, Internet topology analysis has become a popular research area and the analysis can be computationally expensive for such a large scale network. In this paper, we have implemented algorithms to find the shortest paths on large scale Internet topology graphs based on real topology data using breadth-first search. The algorithms have been implemented on graphical processing units (GPUs) using the CUDA platform. We performed our performance measurements on graph sizes ranging from 1,100 to 6.8 million nodes and achieved a maximum speed up of 47x on a single GPU and 124x speed up using 8 GPUs for 100 different starting points.