GraphVine: A Data Structure to Optimize Dynamic Graph Processing on GPUs

Abstract: Graph processing on GPUs is gaining momentum due to the high throughputs observed compared to traditional CPUs, attributed to the vast number of processing cores on GPUs that can exploit parallelism in graph analytics. This paper discusses a graph data structure for dynamic graph processing on GPUs. Unlike static graphs, dynamic graphs mutate over their lifetime through vertex and/or edge batch updates. The proposed work aims to provide fast batch updates and graph querying without consuming too much GPU memory. Experimental results show improved initialization timings by 1968-1269024%, improved batch edge insert timings by 30-30047%, and improved batch edge delete timings by 50-25262% while consuming less memory when the batch size is large.