20. – 25. Mai 2012, Dagstuhl Perspectives Workshop 12212
Co-Design of Systems and Applications for Exascale
Auskunft zu diesem Dagstuhl Perspectives Workshop erteilt
With Petascale computing being a reality today, the focus of the computational science community is already on the next barrier -- exascale computing. With systems even more powerful by orders of magnitude, scientists start thinking about the possibilities and challenges. This workshop addressed the many scientific, technological, and financial challenges of exascale level computing with the hypothesis that exascale computing is only possible by co-designing across different levels of software, hardware, and the surrounding infrastructure.
The workshop program has been composed of a series of short talks, less than 20 minutes on average, and extensive time for discussions. Starting with an overview of the workshop motivation and the general methodologies for co-design, different aspects of co-design have been addressed. This has been followed by talks on modeling, simulation and tools, as well as programming models, runtime support and compilers. The second part addressed the specific problems of system-software for performance, power and reliability and the resulting system architectures, while finally application level aspects of exascale co-design have been discussed between the participating experts from different areas of high performance computing. In all discussions it has been important to tackle a multidimensional combination of major challenges associated with the development of exascale systems and applications from different angles instead of addressing an isolated aspect.
The results of the workshop are manifold (A scientific paper will be created within the next months): Firstly, the vision based on the requirements of the scientific community is thus "to provide exascale capabilities to scientific and engineering applications", where it is important to notice that exascale means extreme scale or large scale, not the particular barrier of exaflop performance looming ahead. With this vision at hand, the participating experts identified their particular role and mission as follows: "to co-design systems such that they reach exascale capabilities within the given technological and non-technical (social,...) boundaries". Each expert has been knowledgeable on a distinct layer of the exascale architecture, the mission requires expertise across all layers, and exascale computing requires involvement from all relevant areas of computer science in order to perform exascale co-design of hard- and software, including also different levels of software working closely together with hardware and the interfacing to the environmental infrastructure. This has lead to the definition of co-design, where two or more distinct activities collaborate on and across different layers to design a system architecture for a specific goal.
In summary, the workshop has reflected on the current state of petascale machines providing multiple examples from world-leading machines and using them to derive the barriers on the road towards exascale computing. Looking beyond the current research into the future, where exascale computing will become feasible, we have been trying to identify the exascale roadmap with intermediate goals and pitfalls on the way to exascale, and leveraging the combined forces of computer science to overcome them.
- Software Engineering
- Power Efficiency