September 10, Thursday
Emerging Exascale Problems and Solutions
In the past 12-24 months, new research problems have emerged in the community preparing for the Exascale systems that go beyond the classic Power, Resilience and Scalability issues identified long ago and for which solutions (at least partial ones) have been identified. The objective of this panel is to introduce the new and emerging problems, present research directions and discuss potential solutions.
The panelists are top researchers, well known for their contributions in HPC and the Exascale landscape:
Panelists:
- Bronis de Supinski, Lawrence Livermore National Laboratory, USA
- Beth Plale, Indiana University, USA
- Satoshi Matsuoka, Titech, Japan
- Taisuke Boku, U. Tsukuba, Japan
- Xian-He Sun, IIT, USA
The moderator will be Franck Cappello, ANL, University of Illinois at Urbana Champaign and Inria.
The panelists will cover critical issues in four major areas:
- Hardware:
- How will technology evolve leading up to 2023?
- Will the two architecture tracks envisioned for CORAL survive?
- Are there unexpected large changes ahead, e.g., no more DRAM?
- Plateauing of I/O bandwidth/Deep hierarchical memory:
- Is it really the trend?
- Can in-situ data analytics solve the problem?
- Can lossy compression be used for all I/O needs (including checkpoint/restart?)
- Do we need new memory movement optimization techniques?
- Do we need new APIs to manage data movements in this deep memory hierarchy?
- Convergence of HPC and Big Data/Connection to experiments:
- Will convergence happen and if so, what are the use cases?
- What will the hardware look like?
- What needs to change in the software stack?
- Does virtualization have a future in HPC?
- Security and Trust:
- What are the known techniques to improve data/computation trustworthiness? Can we integrate them into HPC?
- Can we detect corruptions from insiders?
- Do Exascale HPC systems open new cyber security problems?
The format will be a 10 minute talk for each panelist followed by 20 minutes of questions from the audience.