Case Study FAQ
What is NERSC?
NERSC is the National Energy Research Scientific Computing Center, the high-end scientific computing facility for the Department of Energy's Office of Science. NERSC provides a unique service, serving about 4,000 scientists and about 400 research projects spanning all science areas of interest to the Office of Science.
How are NERSC resources allocated to scientists?
DOE program managers allocate NERSC resources to scientific research teams. Researchers apply for resources through the Energy Research Computing Allocations Process (ERCAP).
Why is NERSC collecting computational requirements?
NERSC is collecting computational requirements from its users to help the Office of Science and NERSC plan for the capability, capacity, technology, and services required to meet Office of Science objectives. Information obtained is used to guide procurements, staffing, and to improve the effectiveness of NERSC computational services. Given that new large-scale systems generally require several years to deploy, we must collect requirements several years in advance.
Who is NERSC asking about requirements?
NERSC is asking for input from its major stakeholders. These include, but are not limited to:
- DOE Office of Science Program Managers;
- Authoritative representatives from major research programs;
- Research and technical staff from national laboratories, universities, supercomputer centers, and the like;
- Key individuals at non-DOE institutions with significant DOE collaborators.
What is a case study?
A case study is a narrative that provides a description of the scientifc goals and the importance of the research, a description of how the science problems drive the computational work, a statement of the computational resources that would be required to meet the science needs, and a description of how all of this may change within a 5-year timeframe.
What kind of requirements information is of interest to NERSC?
NERSC and DOE want to collect information regarding computing, storage, networking, visualization and support needs. There are two ways to look at the issue. One is to answer the question "what level of NERSC facilities and support are required to carry out your scientific goals in the next 3-5 years?" The other is to consider what additional science you would do if you were given an allocation 10x, 100x, and 1000x larger than today (using today's ERCAP award as reference).
We are interested in algorithmic-level information because it helps NERSC understand what kind of computer architectures and which key system attributes might best serve the science. We are interested in such characteristics as typical problem sizes and processor counts, memory needs, data migration needs, storage needs and the like. Another important dimension for today's computer systems is parallelism - how the simulation codes express it and science features enhance or limit it. Still another dimension is how simulations are actually run and the scientific purpose of the runs. Possibilities include testing/verifying new codes as opposed to long-term simulations as opposed to time-critical runs.
Many scientists working at NERSC don't know exactly all the answers to these questions - that's OK. Case study authors should do their best to describe their project's needs and if we need clarification we'll ask for it.
Historical trends are included in this chart for reference. Also plotted are the results from the first round of requirements workshops, 2009-2011.
Do I need to be a computer science expert to provide requirements?
No. NERSC is not asking that requirements be submitted in detailed computer science terms. Rather, case study authors should try to explain in terms that are familiar to them the science that they are doing and the ways in which their research drives computational needs.
If clarification is needed, NERSC will ask for more information. However, the key idea is that NERSC is trying to elicit requirements from the user community from the users' perspective rather than asking physicists, chemists and materials scientists to become computer experts in order to communicate their needs.
How long is a case study?
About a page and a half. It consists of five main sections:
- Summary and Scientific Objectives (1-2 paragraphs)
- Methods of Solution (description of codes, algorithms)
- HPC Requirements (includes a table summarizing)
- Support Services and Software Required
- Emerging HPC Architectures and Programming Models (describe your strategy for computing in such a highly parallel, multi-core environment.