Enhancing HPC Application Performance Based on Graph Computation

HPC systems provide the computational power and parallelism needed to handle complex and large-scale graph computations efficiently. Graph computations are inherently parallelizable because many graph algorithms can be divided into independent tasks that run concurrently. HPC systems, which are designed for parallel processing, can leverage this parallelism to speed up graph computation. Graph computation, in turn, can enhance HPC by improving resource utilization and optimizing data communication. For instance, by structuring data in a graph format, it becomes easier to optimize data locality and reduce access latency, thereby enhancing overall computation efficiency. This talk will explore the effectiveness and potential of graph analytics across different parameter settings and its impact on various downstream HPC and AI applications, including node classification and link prediction.