Abstract: A growing disparity between simulation speeds and I/O rates makes it increasingly infeasible for applications to save all results for analysis. In this new world, applications must increasingly perform online data analysis and reduction—tasks that introduce algorithmic, implementation, and programming model challenges that are unfamiliar to many scientists and that have major implications for the design of various elements of exascale systems.
This trend has spurred interest in online data analysis and reduction methods, motivated by a desire to conserve I/O bandwidth, storage, and/or power; increase accuracy of data analysis results; and/or make optimal use of parallel platforms, among other factors. This requires our community to understand a clear yet complex relationships between application design, data analysis and reduction methods, programming models, system software, hardware, and other elements of a next-generation High Performance Computer, particularly given constraints such as applicability, fidelity, performance portability, and power efficiency.
There are at least three important topics that our community is striving to answer: (1) whether several orders of magnitude of data reduction is possible for exascale sciences; (2) understanding the performance and accuracy trade-off of data reduction; and (3) solutions to effectively reduce data while preserving the information hidden in large scientific data. Tackling these challenges requires expertise from computer science, mathematics, and application domains to study the problem holistically, and develop solutions and hardened software tools that can be used by production applications.