CRI: IAD Exploiting Multicore Processor Technology for Interactive Supercomputing

该项目团队将在明尼苏达大学计算科学工程实验室（LCSE）建立一个系统，专门针对通过交互式超级计算实现探索性科学。 基于Cell和基于Intel的多核CPU技术都将应用于Infiniband集群，以实现团队的持续性能水平？的应用程序代码接近2 Tflop/s。 多核CPU的原始处理能力将与快速连接的磁盘系统、最新的GPU技术和LCSE的PowerWall显示器相结合，通过PowerWall上用户控制的流动可视化，实现对正在进行的流体动力学模拟的交互式控制和即时反馈。 该系统还将用于基因组学的应用。 该项目将支持天体物理学和地球物理学流体动力学，基因组学，科学计算数值算法的多核实现，超大型数据集的高性能科学可视化，网格计算和作业/资源调度的研究。 它还将支持教学和学生培训。 该项目涉及LCSE团队与丰迪拉克部落和社区学院的持续合作。 使用新的交互式超级计算系统开发的技术和软件应该在高性能计算界具有广泛的适用性。 该软件将部署在国家中心的超级计算系统上，在更面向生产的环境中，可以在更高的网格分辨率下享受启用的交互能力。

The project team will build at the University of Minnesota's Laboratory for Computational Science & Engineering (LCSE) a system especially targeted at enabling exploratory science through interactive supercomputing. Both Cell-based and Intel-based multicore CPU technology will be applied in an Infiniband cluster to achieve sustained performance levels with the team?s application codes approaching 2 Tflop/s. The raw processing power of the multicore CPUs will be combined with fast attached disk systems, the latest GPU technology, and the LCSE's PowerWall display to enable interactive control of on-going fluid dynamics simulations with immediate feedback through user-controlled flow visualizations on the PowerWall. The system will also be utilized for applications in genomics. This project will support research in astrophysical and geophysical fluid dynamics, genomics, multicore implementation of numerical algorithms for scientific computation, high performance scientific visualization for very large datasets, Grid computing, and job/resource scheduling. It will also support teaching and student training. The project involves an on-going collaboration of the LCSE team with the Fond du Lac Tribal and Community College. The techniques and software developed using the new interactive supercomputing system should have broad applicability in the high performance computing community. This software will be deployable on super-computing systems at national centers, where the enabled interactive capability can be enjoyed at much higher grid resolutions in a more production-oriented context.