A Partitioning Scheme for Large Scale Clique Counting on Single GPU

With the proliferation of graph structure data, mining patterns in graphs, especially clique counting, have become important for numerous applications, such as graph anomaly detection and community detection. Hence, improving the performance of clique counting is crucial. For many years, researchers have been trying to accelerate by leveraging the latest trends in processor architectures, such as multicores and many-core processors like GPU. While there are several GPU-based general graph mining frameworks [1] exist, the specific clique counting problem has been less explored on GPUs [2] . Moreover, as the real-world graphs have been growing rapidly, the existing frameworks do not support the graphs that do not fit to the capacity of single GPU memory.