Aperture

Abstract

One of the most powerful ways to explore data is to visualize it. Visualizations underpin data wrangling, feature space explorations, and understanding the dynamics of phenomena. Here, we explore interactive visualizations of voluminous, spatiotemporal datasets. Our system, Aperture, makes novel use of data sketches to reconcile I/O overheads, in particular the speed differential across the memory hierarchy, and data volumes. Queries underpin several aspects of our methodology. This includes support for a diversity of queries that are aligned with the construction of visual artifacts, facilitating their effective evaluation over the server (distributed) backend, and generating speculative queries based on a user's exploration trajectory. Aperture includes support for different visual artifacts, animations, and multilinked views via scalable brushing-and-linking. Finally, we also explore issues in effective containerization to support visualization workloads. Our empirical benchmarks profile several aspects of visualization performance and demonstrate the suitability of our methodology.