Scalable and High Performance Betweenness Centrality on the GPU