Faodail: Enabling In Situ Analytics for Next-Generation Systems
Workshop: ISAV 2017: In Situ Infrastructures for Enabling Extreme-Scale Analysis and Visualization
Authors: Patrick M. Widener (Sandia National Laboratories)
Abstract: Existing approaches for in situ analysis and visualization (ISAV) assume that scientific simulations are written in a bulk synchronous parallel (BSP) execution model. However, because of the projected increase in heterogeneity on future systems, alternative execution models, e.g., asynchronous many-task (AMT), have been proposed. In an AMT environment, application data is no longer fixed to a particular location which makes in situ processing more difficult. One solution to this problem is to enhance the data management services used by AMT runtimes to make their data accessible to both AMT applications and non-AMT ISAV tools. Decoupling these data management services from the AMT runtime provides an opportunity to support ISAV tools that can interact with either AMT or BSP applications.
In this paper, we introduce Faodail, a new data management layer that can support the diverse needs of multiple communities on modern platforms. While Faodail is designed to serve as a native data management service for Sandia's DARMA AMT framework, it also provides a flexible means of integrating applications with ISAV tools. We explore this idea with an example that connects ISAV tools to a particle-in-cell plasma simulation.