Prof. Sathish Vadhiyar’s lab at Department of Computational and Data Sciences (CDS) employed a novel divide-and-conquer paradigm for simultaneous formation of communities on the different parts of the graphs on the CPU and GPU cores on SahasraT GPU nodes.
This project, executed in 2019 by Anwesha Bhowmick, a student member at the lab, proposed a novel hybrid CPU-GPU algorithm for large-scale community detection in graphs. A divide-and-conquer strategy was followed where communities are formed on different parts of the graph simultaneously on CPU and GPU. Subsequently, the doubtful vertices are resolved during the merging step. The work was extended to multiple nodes of CPU and GPU, where a novel merging step is employed.
“SahasraT provides a multi-GPU node environment that helps explore billion-vertex problems that can otherwise not be explored on lab machines. Our experiments with multiple nodes show good scalability of the multi-node multi-GPU algorithm with increasing number of nodes.”, says Prof. Vadhiyar
Publications
Anwesha Bhowmik, Sathish Vadhiyar: HyDetect: A Hybrid CPU-GPU Algorithm for Community Detection. HiPC 2019: 2-11.