Date of Award:
12-2026
Document Type:
Dissertation
Degree Name:
Doctor of Philosophy (PhD)
Department:
Computer Science
Committee Chair(s)
Steve Petruzza
Committee
Steve Petruzza
Committee
John Edwards
Committee
Soukaina Filali Boubrahimi
Committee
Yang Shi
Committee
Som Dutta
Abstract
Scientists use supercomputers to simulate weather, blood flow, plasma, and other systems. The largest machines produce more data than can be saved to disk, so scientists are only able to analyze a fraction of what gets generated. Visualizing the data during the run slows the simulation, because the rendering shares the simulation’s graphics processing units (GPUs). Most simulation codes offer no way to interact with a run in progress. The results are often too large to visualize on a local workstation, and exploring them on the supercomputer means waiting in the queue and spending allocation time. We call these three problems the interactivity gap. This dissertation makes five contributions toward closing it.
The first, a “Free Lunch” framework, moves rendering to the central processing unit (CPU) cores, which sit idle while the simulation runs. In our largest tests, this cut run time by 21 to 40 percent compared with GPU rendering. The second, now part of Ascent, lets scientists pause a running simulation, inspect or change it, and continue. The third brings the same controls to a web browser. The fourth stores results as compact sets of 3D Gaussians that a single GPU can render with any transfer function, the mapping from values to color and opacity. The fifth splits larger datasets across up to 256 GPUs. Each GPU independently trains 3D Gaussians for its own part, so it scales by adding GPUs.
Interacting with a running simulation instead of restarting it, and exploring results locally, both save allocation time on a shared supercomputer. The steering tools allow only the changes that a simulation’s developers chose to support, and the compact models take time to train and only approximate the original data.
Recommended Citation
Sewell, Andres, "Closing the Interactivity Gap: Towards Efficient In Situ Analysis, Bidirectional Steering, and Gaussian-Based Visualization for Large-Scale Scientific Computing" (2026). All Graduate Theses and Dissertations, Fall 2023 to Present. 948.
https://digitalcommons.usu.edu/etd2023/948
Included in
Copyright for this work is retained by the student. If you have any questions regarding the inclusion of this work in the Digital Commons, please email us at .