Accurate exploratory analysis and visualization of time-varying flow data is challenging on modern day supercomputers. Although CFD simulations can be performed at high resolutions and large amounts of data can be generated, the existing limitations in I/O performance prevent scientists from saving data at high temporal resolutions, i.e., data is stored infrequently. As a result, the traditional paradigm for flow visualization, which is based on the Eulerian paradigm, performs poorly. In this presentation, I will show how using a combination of in situ processing and the Lagrangian representation of time-varying flow field information can significantly increase accuracy and/or reduce storage requirements. In particular, I will focus on establishing viability within in situ constraints, as well as best practices for the Exascale computing era.
Posted by: Steve Petruzza