Importance-driven In situ Analysis and Visualization

The advent of exascale has enhanced the computing capacity to unprecedented scales. Scientific applications now generate massive amounts of data in a few seconds. However, improvement in memory, I/O and network bandwidth has been sub-exponential, resulting in an increasing gap between the rate at which data may be generated and consumed. Data is typically analyzed and visualized after simulations. In situ processing implies analyzing/visualizing data as soon as it is generated, often bypassing the disk I/O bottleneck. Analyzing every time step will increase the end-to-end simulation-analysis time. However most works determine the frequency of analysis/visualization without examining the data content. This may result in omission of critical time steps of the simulations. We propose improving the simulation-analysis-visualization workflow time considering the importance of data. We monitor the changes in data in an ongoing simulation and transfer only the most significant time steps thereby further reducing the data transfer time (by 68%), which is often a bottleneck for in situ analysis.