Computing at Scale
Thursday’s examination of the large scale computations and technologies needed to support fusion science in Simulation at the Petascale and Beyond for Fusion Energy Science falls into both the Computing at Scale and Application Horizons themes and provides a view into one of the most pressing research areas of our time.
Big compute usually means big data, and data management is receiving increased focus as petabyte-sized data stores become commonplace. The BitDew environment is discussed in the Visualization and Data Management session after lunch, highlighting the features that make it an interesting solution for automatic and transparent data management on computational grids.
Finally, the panel Disruptive Technologies: Weapons of Mass Disruption looks at the potential accelerative effects of quantum computing, flash storage, cheap and low power optical communications, and 3D chip stacking with an eye toward the possible goal of reaching an exascale system by 2020.
Both the physical and software infrastructure needed to support supercomputing get a lot of attention on Thursday. Proactive Process-Level Live Migration in HPC Environments examines the use of health monitoring to anticipate node failures before they happen, and then transparently migrate processors away from those nodes before a computation is interrupted.
The Grid Virtualization and Overlays papers session early in the afternoon presents papers on the use of overlays for efficient wide-area data transfers, examines the cost of doing science in the cloud, and using server-storage virtualization for integration and load balancing in datacenters.
Thursday also looks at the physical side of computation support infrastructure. Zero-emission Datacenters: Concept and First Steps compares three approaches to cooling datacenters and provides evidence that a novel approach studied by the authors cuts energy use by almost a factor of two.
Moving from Data Center Efficiency to Data Center Productivity looks at the evolution of attitudes as we move from managing power and cooling for efficiency and learn to manage datacenters for productivity. Finally, the Masterworks session Save Energy Now in Computer Centers presents the practical steps and real world results that Lawrence Berkeley National Lab has experienced on its path to more responsible energy use in its datacenters.
HPC and the Energy Challenge: Issues and Opportunities features Jeffrey Wadsworth and a discussion of how petascale computing will enable a revolution in the way we provide for the energy needs of a growing world population.