曙光4000A上三维叠前深度偏移并行计算的应用设计

波动方程叠前深度偏移在地震勘探成像处理方面起着不可替代的作用。随着高性能大规模并行计算机技术的发展,波动方程叠前深度偏移计算在地震勘探中的应用有了很大进步。在波动方程叠前深度偏移处理中,庞大的数据规模与海量计算对计算性能提出了很高的要求。曙光4000A超级计算机系统是我国目前峰值速度最快的商用超级计算机系统,无论是硬件平台建设还是应用软件的配置方面,都具有良好的应用性能。基于该系统设计的三维波动方程叠前深度偏移(炮域)PSDM软件,采用动态负载平衡并行计算模式,具有较高的计算效率,高度的可扩展性和可靠性。     

超算环境下的并行切图性能分析

在传统GIS二维地图缓存生成过程中,一般用普通的PC机来生成地图缓存.普通PC机的内存一般不大于8 G,CPU主频一般不大于2.2 GHz,且核数不大于8核,硬盘一般为机械硬盘,容量不大于2 T.受限于硬件性能的瓶颈,切图效率较低,不能够分析出上述各硬件因素中制约二维地图缓存性能的关键性因素.相比之下,超算环境有着丰富的资源,可将上述任一资源延伸到极致,因此,本文借助超算环境,研究分别增大内存、CPU主频、硬盘空间等因素后对地图缓存效率的影响.基于此,本研究在郑州大学超级计算机中心专门进行了比较测试.郑州大学超级计算机中心拥有两种类型的CPU,每个主机拥有128 G内存,CPU分为普通CPU和KNL CPU,分别拥有32核和64核,所有主机共享802 T的磁盘阵列,可以认为对于二维地图缓存来说硬件资源是无限的.本文利用郑州大学超级计算机中心的理想环境分析出了在内存和CPU等因素都不受限的情况下,影响二维地图缓存效率的关键因素.     

BCCAGCM模式在神威·太湖之光系统的优化

开展气象数值模式在神威·太湖之光系统的移植与优化,对研究模式与新型计算架构的适应性有重要意义。该文以BCCAGCM模式为研究对象,将其移植到神威·太湖之光全国产异构众核计算系统,进行性能分析,对模式动力框架和物理过程计算结构进行调整,将计算核心段采用OpenACC技术进行众核加速优化,大量代码进行算法重构。结果表明:各核心段计算效率基本达到未优化的3倍左右,最高可达14倍左右,将各核心段集成,形成异构众核集成版本,可正确、稳定运行,计算误差合理。在不同并行规模,采用从核对模式整体计算进行加速效果比较稳定,基本保持在1.9倍,26000核并行规模动力试验并行效率约70%,其他试验约为57%。     

Toward an integrated framework for geosensor grid

Abstract

Grid computing is deemed as a good solution to the digital earth infrastructure. Various geographically dispersed geospatial resources can be connected and merged into a ‘supercomputer’ by using the grid-computing technology. On the other side, geosensor networks offer a new perspective for collecting physical data dynamically and modeling a real-time virtual world. Integrating geosensor networks and grid computing in geosensor grid can be compared to equipping the geospatial information grid with ‘eyes’ and ‘ears.’ Thus, real-time information in the physical world can be processed, correlated, and modeled to enable complex and advanced geospatial analyses on geosensor grid with capability of high-performance computation. There are several issues and challenges that need to be overcome before geosensor grid comes true. In this paper, we propose an integrated framework, comprising the geosensor network layer, the grid layer and the application layer, to address these design issues. Key technologies of the geosensor grid framework are discussed. And, a geosensor grid testbed is set up to illustrate the proposed framework and improve our geosensor grid design.     

网格计算核心技术浅析

介绍了网格系统的特点和几种主要的网格体系结构,分析了网格系统中高速通信、安全、资源管理、任务管理与任务调度等核心技术.针对资源管理技术在网格系统中的重要性,作了较为详细的讨论并且提出了一种类似于搜索引擎机制的方法来组织网格资源,构造局部网格系统,各局部网格系统可以通过P2P技术互连,从而建立更大的网格系统,实现系统的可扩充性.     

The Lofar Central Processing Facility Architecture

Reconfiguration is a key feature characteristic of the LOFAR telescope. Software platforms are utilised to program out the required data transformations in the generation of scientific end-products. Reconfigurable resources nowadays often replace the hard-wired processing systems from the past. This paper describes how this paradigm is implemented in a purely general-purpose telescope back-end. Experiences from high performance computing, stream processing and software engineering have been combined, leading to a state-of-the-art processing platform. The processing platform offers a total processing power of 35 TFlops, which is used to process a sustained input data- stream of 320 Gbps. The architecture of this platform is optimised for streaming data processing and offers appropriate processing resources for each step in the data processing chains. Typical data processing chains include Fourier transformations and correlation tasks along with controlling tasks such as fringe rotation correction. These tasks are defined in a high level programming language and mapped onto the available resources at run time. A scheduling system is used to control a collection of concurrently executing observations, providing each associated application with the appropriate resources to meet its timing constraint and give the integrated system the correct on-line and off-line look and feel.     

3D finite-difference simulations of strong ground motions in the Yanhuai area, Beijing, China during the 1720 Shacheng earthquake (Ms 7.0) using a stochastic finite-fault model

Three-dimensional finite-difference (3D-FD) simulations of earthquake wave propagations in the Yanhuai area were performed for the 1720 Shacheng earthquake (M_s 7.0) using a stochastic finite-fault model, running on a parallel supercomputer Hitachi-SR8000 (http://www.lrz-muenchen.de). A stochastic finite-fault model was implemented into the 3D-FD program. The basic idea of the stochastic finite-fault model is that the fault plane can be subdivided into several subfaults (or elements, sources). Radiation from a large earthquake is the sum of contributions from all subfaults with proper time delays, each of which acts as a small independent double-couple point source. Heterogeneity of the fault rupture process was modeled by randomizing the location of initial rupture (hypocenter), slip vectors (slip, rake), and rise-times of subfaults in this study. A 3D velocity model of the Yanhuai area was constructed based on studies that analyzed available geological and geophysical information. A grid increment of 75 m in three directions was used in the 3D-FD simulation, which made it possible to capture the short period information with a resolution as low as 0.5 s in sediment regions. The uncertainties of simulated results caused by the stochastic finite-fault model were studied with a homogeneous 3D model. We found that the effects of the randomness of source on simulated ground motions are only limited in near-fault-region including the surface exposure of the fault and its vicinities, which occupies about 5% of the whole study area. This article presents an integrated approach for simulating the strong ground motions for engineering purpose using the 3D-FD method. Such simulations would be useful for hazard mapping, land using planning, insurance rate assessment, particularly in planning, preparedness, and coordinating emergency response, which must be based on realistic situations induced by concrete (historic or scenario) earthquakes.     

BCC_AGCM大气环流模式异构众核加速技术

针对未来高性能计算在CPU混合架构上的发展趋势,本文对大气环流模式BCC_AGCM中的部分核心段在混合架构的神威·太湖之光高性能计算机系统上进行众核加速分析与优化。文中以核心段quad为例,通过对代码内容结构重写和使用OpenACC语言对程序进行运行指示的方式,将该核心段成功移植加速并通过了结果正确性验证。结果表明,使用OpenACC语言能够以对源代码的较小改动来获得一定的加速比,在异构环境中源代码优化加速提升的效果取决于程序结构、循环并行颗粒度以及地址连续性。     

Optimal orientations of discrete global grids and the Poles of Inaccessibility

ABSTRACT

Spatial analyses involving binning often require that every bin have the same area, but this is impossible using a rectangular grid laid over the Earth or over any projection of the Earth. Discrete global grids use hexagons, triangles, and diamonds to overcome this issue, overlaying the Earth with equally-sized bins. Such discrete global grids are formed by tiling the faces of a polyhedron. Previously, the orientations of these polyhedra have been chosen to satisfy only simple criteria such as equatorial symmetry or minimizing the number of vertices intersecting landmasses. However, projection distortion and singularities in discrete global grids mean that such simple orientations may not be sufficient for all use cases. Here, I present an algorithm for finding suitable orientations; this involves solving a nonconvex optimization problem. As a side-effect of this study I show that Fuller's Dymaxion map corresponds closely to one of the optimal orientations I find. I also give new high-accuracy calculations of the Poles of Inaccessibility, which show that Point Nemo, the Oceanic Pole of Inaccessibility, is 15?km farther from land than previously recognized.     

近30年中国天气预报业务进展

介绍了我国天气预报技术的发展现状，指出在大气探测技术、数值预报及可视化技术发展的推动下我国传统的天气预报向定量、定时方向发展，并系统阐述了天气预报业务发展过程和存在的问题，最后提出了对我国天气预报业务发展的思路。