The three DOE Office of Advanced Scientific Computing (ASCR) High Performance Computing (HPC) Facilities – NERSC, the Argonne Leadership Computing Facility (ALCF), and the Oak Ridge Leadership Computing Facility (OLCF) – are conducting a series of six Exascale Requirements Requirements Reviews to identify mission-critical computational science objectives within the DOE Office of Science (SC) in 2020 and 2025. These reviews will help assure that ASCR facilities will be able to meet SC needs through the specified time frames.
The reviews, one for each of the six SC program offices will bring together key computational domain scientists, DOE planners and administrators, and experts in computer science and applied mathematics to determine the requirements for an exascale ecosystem that includes computation, data analysis, software, workflows, HPC services, and whatever else is needed to support forefront scientific research.
The review will cover and collect the mission science goals for each office and then determine the requirements of the exascale ecosystem needed to support them. The compute facilities will provide the best picture of forthcoming architectural swim lanes on the path to exascale and baseline programming environments. Requirements will reflect how research groups and code developers will use those architectures. Requirements include those related to computation; data movement, storage, and analysis; programming approaches; training; and other support services. The reviews will identify requirements in software, services, and hardware that might not be met by the current facilities plans but might be filled by ASCR research efforts and the Exascale Computing Initiative.
The final deliverable will be a report for each Office detailing the review findings. This report will help ASCR Facilities Division develop a strategic roadmap to inform future ASCR compute and data facility investments, ASCR Facility Division partnerships with Office of Science stakeholders (ASCR R&D, BES, BER, FES, HEP, and NP) and inform the Office of Science research agenda.
The process will be defined and implemented by a steering committee consisting of representatives from the three ASCR HPC Facilities. Steering committee members are
- Katie Antypas, NERSC
- Sudip Dosanjh, NERSC
- Richard Gerber, NERSC
- James Hack, OLCF
- Paul Messina, ALCF
- Katherine Riley, ALCF
- Tjerk Staatsma, OLCF
- Jack Wells, OLCF
- Tim Williams, ALCF
Each review will have a program committee, including a chair and co-chair who are leaders from the scientific community, and members of the steering committee. Attendees will be chosen in consultation with the committee chair, co-chair and DOE SC program managers. The goal is to limit the number attendees to about 50 participants including domain scientists, experimental facilities, engaged applied mathematicians and computer scientists, software developers. Attendance is by invitation only. The reviews will be held in the DC area so that DOE SC program managers can observe.
The workshop chair and the steering committee will select the themes and objectives for the breakout sessions to be held at the workshop and identify the breakout leads. Breakout session leads will be invited based on their technical expertise and vision for the computational needs for the scientific field, their ability to focus on the key technical issues, and their skills and experience in writing an extended reporting document that will serve the intended audience. Each breakout session should have a dedicated theme. With a target of 50 technical participants and an optimal number of 10-15 for each breakout, the workshop should limit the number of breakout sessions to four. Some ASCR research-funded mathematicians and computer scientists will be invited. The breakout leads will be responsible for putting together a discussion draft whitepaper before the review. The breakout leads will also identify areas/applications to target for case studies. The program committee will send out a template for documenting the case study requirements well before the review. The case studies will be discussed in the breakout sessions and will feed directly into the white papers for each breakout session.
The reviews will be administered by the facilities on a rotating basis as shown above.