In situ generated probability distribution functions for interactive post hoc visualization and analysis

• Published in: 2016 IEEE 6th Symposium on Large Data Analysis and Visualization (LDAV) pp. 65 - 74
• Main Authors: Ye, Yucong Chris, Neuroth, Tyson, Sauer, Franz, Kwan-Liu Ma, Borghesi, Giulio, Konduri, Aditya, Kolla, Hemanth, Chen, Jacqueline
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2016
• Subjects: Analytical models; Computational modeling; Data models; Data visualization; G.3 [Mathematics of Computing]: Probability and Statistics - Distribution Functions; H.3.2 [Information Systems]: Information Storage and Retrieval - Information Storage; Histograms; I.6.6 [Computing Methodologies]: Simulation and Modeling - Simulation Output Analysis; J.2 [Computer Applications]: Physical Sciences and Engineering - Physics; Probability density function; Probability distribution
• DOI: 10.1109/LDAV.2016.7874311

The growing power and capacity of supercomputers enable scientific simulations at extreme scale, leading to not only more accurate modeling and greater predictive ability but also massive quantities of data to analyze. New approaches to data analysis and visualization are this needed to support interactive exploration through selective data access for gaining insights into terabytes and petabytes of data. In this paper, we present an in situ data processing method for both generating probability distribution functions (PDFs) from field data and reorganizing particle data using a single spatial organization scheme. This coupling between PDFs and particles allows for the interactive post hoc exploration of both data types simultaneously. Scientists can explore trends in large-scale data through the PDFs and subsequently extract desired particle subsets for further analysis. We evaluate the usability of our in situ method using a petascale combustion simulation and demonstrate the increases in task efficiency and accuracy that the resulting workflow provides to scientists.