Benchmarking In Situ Triggers Via Reconstruction Error

Abstract

This work considers evaluating in situ triggers using reconstruction error. Our experiments use data from the Nyx and Cloverleaf simulation codes, and focus on two key topics. The first topic aims to increase understanding of total reconstruction error, both with respect to the impact of adding more time slices and with respect to the variation from different time slice selections. The second topic evaluates performance for two current approaches: entropy-based triggers and evenly spaced time slices. Finally, we use these study components to construct a benchmarking system that enables visualization scientists to reason about triggers.