China National Grid Project (CNGrid): Phase II  (2006-2010)

HKU constructed the initial version of the HKU Grid Point with the Gideon cluster installed in 2002 (phase I). CNGrid HKU Grid Point was officially launched in December, 2005. It is developed and managed by the System Research Group (SRG) at The University of Hong Kong (HKU). The goal of this grid point is to construct an easy-to-use, efficient, and flexible application development platform, to support high performance applications, such as financial computing, new drug discovery, data mining, image rendering and animation, etc. At 2005, HKU Grid Point consists of several large clusters including CS Department’s Gideon300 PC Cluster and Computer Center’s HPCPOWER cluster, can provide over 3Tflops computing power in total. Besides supporting high performance computing services to CNGrid project, we also involve in the development of Hong Kong Grid and the Pearl River Delta Region Grid (with SIAT), and are members of Pacific Rim Application and Grid Middleware Assembly (PRAGMA) and European-based e-science project (EGEE).

Figure 1: CNGrid Project (phase II): Grid Node Distribution

The China National Grid II (CNGrid-II) will connect 12 grid nodes across the country, including:

1. 中科院计算机网络信息中心（Computer Network Information Center, CAS）(DeepComp 7000 : 143Tflops)
2. 上海超级计算中心 (Shanghai Supercomputing Center): 魔方Dawning 5000A: 230TFlops
3. 清华大学 Tsinghua University (Beijing) : 探索3号万亿次集群 (校985“三期”投资三千万建设“探索100”集群计算机, 计算性能达到104TFlops)
4. 北京应用物理与计算数学研究所: 银河高性能计算机
5. 中国科技大学 University of China Science and Technology (合肥Hefei, Anhui) 
6. 西安交通大学 Xi’an Jiaotong University: 浪潮天梭TS10000 (3.4T)
7. 甘肃超算中心 曙光5000 系统峰值40Tflops 
8. 山东大学 Shandong U: 浪潮TS10000机群、浪潮TS20000机群
9. 华中科技大学: IBM System p5-575 SMP
10. 中国科学院深圳先进技术研究院: 曙光4000L, 1.5Tflops.
11. 香港大学 Hong Kong University (Hongkong)
12. 国家超算天津中心 : 天河一号A

CNGrid HKU Grid Point

The first phase of the HKU Grid Point project is implemented in early 2010. A 1920 cores of 64-bit Linux Grid point system is installed to support System Research Group (SRG) of Department of Computer Science and Multiple Disciplines Research Project (MDRP) for HKU researchers on Grid computing, with a total processing power, in term of  theoretical peak performance value of 19.43 Tera-FLOPS (and measured Rmax of 14.218 Tera-FLOPS). This system consists of latest hardware such as Intel Nehalem Quadcore CPU, 10Gb Ethernet (10 Gb/s) switch and 4X DDR Infiniband (20 Gb/s) communication network. The second phase of Grid Point project is expected to be completed by early-2011.  It is planned that more computational power based on latest Intel Westmere 6-core Xeon CPUs, latest NVIDIA Fermi Graphics processing unit (GPU) and  fastest communication based on 4X QDR Infiniband (40 Gb/s) communication network will be implemented. It is expected that the theoretical peak performance value of the whole system will be augmented by at least 7.62 Tera-FLOPS after the second phase of implementation. This new HKU Grid Point system is supported by a high speed direct network route between HKU network and China Science and Technology Network (CSTnet) established at late 2009 to provide a dedicated and high-speed network connection to China National Grid. In addition, HKU also has special research network set-up connecting through Internet2, Trans-Eurasia Information Network (TEIN2), Korea Research Environment Open Network (KREONET) and Taiwan Academia Sinica Grid Computing Network (ASGCNET) directly to the research networks of US, Japan, Singapore, Korea and Taiwan for the purpose of fast data transfer between HKU network and other international grid networks.

19.43 Tera-FLOPS Gideon-II cluster (09/2009)		CNGrid HKU Grid Point officially launched in Dec 2005.
		The HKU Gideon 300 cluster: 0.356 Tera-FLOPS (ranked No. 175 on the TOP500 supercomputer list 11/2002)

Research Objectives

Part I: Building HKU Grid Point

• 扩大和升级香港大学网格结点的资源

  • 香港大学网格节点，是香港地区最大处理能力的网格计算资源。它基于戴尔PowerEdge服务器、IBM iDataPlex服务器、英特尔最新的CPU技术以及英伟达Fermi图形处理器技术而构建。第一阶段的港大网格节点计划已于2010年初实现：一个新的拥有1920个内核的64位Linux网格节点系统安装完成，以支持港大计算机系系统研究小组(SRG)和港大多学科研究项目(MDRP)之网格计算。它的理论峰值运算速度达每秒19.43万亿次浮点运算，LINPACK实测运算速度达到14.22万亿次浮点运算。该系统包含最新的硬件，如戴尔PowerEdge服务器，英特尔Nehalem四核心服务器级中央处理器，十万兆以太网交换机和四倍速DDR InfiniBand网卡的通信网络。第二阶段的港大网格节点项目，也已于2011年年初完成：该系统配有最新的英特尔Westmere六核心服务器级中央处理器以及英伟达Fermi图形处理器，以及更快速的四倍速QDR InfiniBand网卡。第二阶段项目对整个系统的理论峰值运算速度扩充了7.62万亿次浮点运算。结合两个阶段的总运算能力，其理论峰值运算速度达到每秒27.05万亿次浮点运算。

• 实现自适应的Java计算环境:
  • Integrating G-JavaMPI2 and JESSICA3
  • JESSICA: a distributed JVM system which allows any multi-thread Java program to be executed in a cluster without code modifications. Up to November, 2008, JESSICA software has been downloaded over 500 times.
  • G-JavaMPI: a Grid middleware supporting large-scale MPI-based parallel Java computing with the support of process migration and checkpointing for dynamic load balancing and fault-tolerance.
• Virtualization Technique for Instant Provisioning of Grid Computing Environment
  • SLIM-VM: a new virtualization technique that supports high-speed dissemination of virtualized Grid computing environment among networked machines.
  • 基于SLIM-VM 的网格部署方案 :
    • 安装操作系统镜像，使用虚拟化，在最大的范围内实现资源的高效共享. 在一个计算结点上同时执行各种截然不同的网格计算任务
•   Smart Grid Connection for Mobile Users  (普适化智能网格接入)
  • 课题涵盖普适计算新环境下的移动计算中间件支持，动态自适应，情境感知，本体论映射，对等网络通信等方面

Part 2: Porting Grid Applications

• 新药发现应用网格 (中國科學院上海藥物研究所)
• Porting Tomcat on JESSICA3. (基于Tomcat的商业Web应用) :
  • Real-time stock quotes (实时股票报价)
  • TPC-W bookstore (网上书店)
• BetterLife 2.0 (Context-aware Recommendation System)
• Constructing HKU Campus Grid using WAVNet