Extreme Heterogeneity Workshop Summary

Event Sponsor: 
ASCR Research, Computer Science Seminar
Start Date: 
Apr 3 2018 - 8:45am
Building/Room: 
Building 240/Room 1404-1405
Location: 
Argonne National Laboratory
Speaker(s): 
Lucy Nowell
Speaker(s) Title: 
Program Manager, ASCR Research, Computer Science

In this talk Dr. Lucy Nowell will introduce herself and her new role within ASCR Research to the audience and describe the goals of the ASCR Extreme Heterogeneity (EH) program.

The Extreme Heterogeneity workshop was held on January 23-25, 2018, in Gaithersburg, MD. Extreme heterogeneity is the result of using multiple types of processors, accelerators and memory/storage in a single computing platform that must support a variety of application workflows to meet the needs of increasingly diverse scientific domains. Extremely heterogeneous supercomputers will be acquired by the ASCR-supported computing facilities as we reach the end of Moore’s Law while still facing rapidly increasing computational and data intensive demands. The ASCR Computer Science research focus for this workshop was on system software, and software development tools and environments for supercomputers that will be delivered for operational use in the 2025-2040 time frame.

The purpose of the EH workshop was to more clearly define the challenges that extreme heterogeneity presents to the software stack and the scientific programming environment and to identify related Computer Science priority research directions that are essential to making extremely heterogeneous systems useful, usable, efficient, and secure for science applications and DOE mission requirements.

The workshop aims were to identify and prioritize research directions by analyzing existing and next-generation computer architectures. The workshop targeted post-exascale architectures including novel technologies that may be developed in the "Post-Moore's Law era" and promising tools and techniques, such as advanced analytics and machine learning, that may enable efficient and productive utilization of such architectures. Participants also discussed options to leverage methods developed by industry, such as approaches to improved developer productivity for Big Data.