SC Conference - Activity Details

Steven W. Schlosser  (Intel Research)

Michael P Ryan  (Intel Research)

Ricardo Taborda  (Carnegie Mellon University)

Julio Lopez  (Carnegie Mellon University)

David R O'Hallaron  (Intel Research)

Jacobo Bielak  (Carnegie Mellon University)

Applications: Models and Analysis

Thursday,  03:30PM - 04:00PM

Room Ballroom E

Large-scale earthquake simulation requires source datasets describing the physical characteristics of the earth in the region under simulation. Physical characteristic datasets are the first stage in the simulation pipeline which includes mesh generation, partitioning, solving, and visualization. In practice, the data is produced in an ad-hoc fashion for each experiment, which has several significant shortcomings. First, a scientist must choose among the available velocity models. Second, repeatedly sampling the model is expensive. Lastly, experiments that use different ground model configurations are difficult, if not impossible, to validate against each other. To address these problems, we propose a new technique in which ground models are fully materialized, and the resulting data efficiently stored in etrees, to be shared among many scientists for many experiments. The constituent ground model is sampled at high resolution and the resulting sample data is reduced by coalescing neighboring samples with homogeneous characteristics into increasingly larger octants.

The full paper can be found in the IEEE Xplore Digital Library and ACM Digital Library