EPIC - An Extract Plug-In Components Toolkit for In-Situ Data Extracts Architecture

Abstract

A new data extract toolkit known as EPIC-Extract Plug-In Components Toolkit has been designed and implemented to create data surface extracts in situ via the TURBO CFD code. New FieldView extracts (eXDB) with a Proper Orthogonal Decomposition (POD) reduced order model were implemented, visualized and analyzed using a new prototype version of the CFD post-processor FieldView. Benchmarks show that the EPIC with POD surface extracts have some memory and computational cost overhead that may be mitigated by the reduced overall user workflow and the overall reduction in data files saved to disk. In addition, the resulting extracts provide new visualization and feature detection capabilities that may be obtained through analysis of the POD modes. This paper presents benchmarks at scale for unsteady Gas Turbine simulations (S-37 and BRI) and explores the feature detection capabilities of POD.

State of the Art and Future Trends in Data Reduction for High-Performance Computing

“…While in situ processing is best performed at the application level of the software stack, middleware and toolkits can increase their performance by facilitating the data extraction. Examples of such frameworks include EPIC [44] and Freeprocessing [46]. A solution based on I/O layer components was proposed in [26].…”

In Situ Methods, Infrastructures, and Applications on High Performance Computing Platforms

“…EPIC [DHH*15] is a toolkit for generating in situ surface extracts and then optionally computing a Proper Orthogonal Decomposition reduced order model for post hoc analysis and visualization. It uses a master‐slave MPI task hierarchy to manage the data from the other analysis and user application tasks.…”