Program

Implementing Exascale Computational Topology

Computational topology has developed over the past three decades to became a standard technique to assist analysis, interpretation and visualization of large-scale data, in particular volumetric data such as generated by numerical simulations. From serial beginnings, work over the last decade under the ECP-Alpine project has delivered tools in vtk-m/viskores that exploit both single-node parallelism and multi-node distribution that now permit processing of single data slices in the TB range, with potential for further scaling. This talk will review the background of the computational topology and its value for data analysis and visualization, then examine the development of these parallel and distributed tools to see how modern hybrid distributed-parallel algorithmic development differs from classical serial development, and look at some parallels with scalable graph algorithms.