The ability to obtain fast I/O using a sustainable, easy-to-use, fast, and portable framework for the largest applications running on high performance computers remains a formidable task. This project researches techniques for two challenges that are faced by the research community in high performance I/O.
1. Fast and scalable I/O that adapts to ever-increasing system complexity in current HPC systems, built with multi-core nodes possessing accelerators and deep memory hierarchies. This research will investigate new techniques to deal with the trends for current-generation supercomputers to use accelerators such as GPGPUs and Intel's Xeon Phi.
2. Sustainable and portable parallel I/O. The project will build on the team's Adaptable I/O System (ADIOS) to develop such an I/O infrastructure that encapsulates research and best practices and can enable scientists to use emerging platforms.
Additional goals are to promote international visibility of the benefits of investment in HPC by demonstrating the capability of ADIOS, and to explore new collaboration mechanisms for international researchers. This research is simultaneously high risk (researching and developing a conceptual solution that can address the I/O challenges faced by large HPC centers when using accelerators) and high reward (this solution will be encapsulated into a sustainable software framework that has a large user base).
