Getting Ready for Exascale Computational Science

Rick Stevens

In this talk I値l outline a possible roadmap to exascale science. I値l discuss some of the scientific problems that may motivate developing an exascale computational capability and some of the challenges in developing applications and systems software for systems that will exploit terascale integration VLSI. I値l discuss the challenges of dramatically increasing levels of concurrency as exascale systems will likely require applications concurrency of 109 a dramatic increase from today痴 largest systems with programmer visible concurrency levels of 105. I値l discuss some of the issues with programming models and I/O as well as the underlying technology challenges such as advanced packaging and power management, optical interconnects, memory architecture and ultra lower power designs for core logic and I/O. Finally I値l discuss the role of future high-end systems in the overall computational ecosystem that includes advanced networking, visualization and analysis systems, highly parallel data architectures and emerging global sensor networks.