Scientists from Sandia National Laboratories, Oak Ridge National Laboratory, the University of Illinois, the University of California Berkeley, and the University of Tennessee have been awarded funding to create a DOE Joint Math/CS Institute.
The Extreme-scale Algorithms and Software Institute (EASI) will focus on closing the application-architecture performance gap by developing architecture-aware algorithms and libraries, and the supporting runtime capabilities to achieve scalable performance and resilience on heterogeneous architectures. Closing the application-architecture performance gap will require both innovative and collaborative research in applied math and computer science.
EASI will leverage the IAA algorithms project and engage the community to:
- Study and characterize the application-architecture performance gaps that can be addressed in the near-term and identify architecture features that future systems may want to incorporate.
- Develop multi-precision and architecture-aware implementations of Krylov, Poisson and Helmholtz solvers, and dense factorizations for heterogeneous multi-core systems.
- Explore new methods of algorithm resilience, and develop new algorithms with these capabilities.
- Develop runtime support for adaptable algorithms that must deal with resilience, scalability, and performance.
- Demonstrate architecture-aware algorithms in full DOE applications on large-scale DOE architectures.
- Distribute the new algorithms and supporting runtime capabilities through widely used software packages.
- Establish a strong outreach program to disseminate results, interact with colleagues and train students and junior members of our community.
EASI will host annual workshops on the latest architecture-aware algorithm technologies. A strong outreach program will fund graduate student research efforts and postdocs. The workshops will be organized to disseminate the research to the larger community so that they can apply the techniques to other algorithms and to facilitate incorporation of the new solvers into other applications.
Top of page