Johnston W E, Jacobson V L, Loken S C, Robertson D W, Tierney B L
Lawrence Berkeley Laboratory, University of California, Berkeley.
Crit Rev Biomed Eng. 1992;20(5-6):315-54.
The next several years will see the maturing of a collection of technologies that will enable fully and transparently distributed computing environments. Networks will be used to configure independent computing, storage, and I/O elements into "virtual systems" that are optimal for solving a particular problem. This environment will make the most powerful computing systems those that are logically assembled from network-based components and will also make those systems available to a widespread audience. Anticipating that the necessary technology and communications infrastructure will be available in the next 3 to 5 years, we are developing and demonstrating prototype applications that test and exercise the currently available elements of this configurable environment. The Lawrence Berkeley Laboratory (LBL) Information and Computing Sciences and Research Medicine Divisions have collaborated with the Pittsburgh Supercomputer Center to demonstrate one distributed application that illuminates the issues and potential of using networks to configure virtual systems. This application allows the interactive visualization of large three-dimensional (3D) scalar fields (voxel data sets) by using a network-based configuration of heterogeneous supercomputers and workstations. The specific test case is visualization of 3D magnetic resonance imaging (MRI) data. The virtual system architecture consists of a Connection Machine-2 (CM-2) that performs surface reconstruction from the voxel data, a Cray Y-MP that renders the resulting geometric data into an image, and a workstation that provides the display of the image and the user interface for specifying the parameters for the geometry generation and 3D viewing. These three elements are configured into a virtual system by using several different network technologies. This paper reviews the current status of the software, hardware, and communications technologies that are needed to enable this configurable environment. These interdependent technologies include: (1) user interface and application program construction methodologies, (2) the interprocess communication (IPC) mechanisms used to connect the software modules of the application, (3) the network protocols and interface hardware used by the IPC for communicating between modules running on separate and independent computing system elements, (4) the telecommunications infrastructure that provides the low-level data transfer functions for the networks that connect the geographically distributed elements used by the application, and (5) the nature of the functional elements that will be connected to form virtual systems.
在未来几年中，一系列技术将逐渐成熟，从而实现完全透明的分布式计算环境。网络将用于将独立的计算、存储和输入/输出元素配置成“虚拟系统”，这些虚拟系统对于解决特定问题而言是最优的。这种环境将使最强大的计算系统成为那些由基于网络的组件逻辑组装而成的系统，并且还将使这些系统面向广大用户群体。预计在未来3至5年内将具备所需的技术和通信基础设施，我们正在开发和演示原型应用程序，以测试和运用这个可配置环境中当前可用的元素。劳伦斯伯克利实验室（LBL）的信息与计算科学部以及研究医学部与匹兹堡超级计算中心合作，展示了一个分布式应用程序，该程序阐明了使用网络配置虚拟系统的问题和潜力。此应用程序通过使用基于网络的异构超级计算机和工作站配置，实现对大型三维（3D）标量场（体素数据集）的交互式可视化。具体的测试案例是对3D磁共振成像（MRI）数据进行可视化。虚拟系统架构由一台从体素数据执行表面重建的Connection Machine - 2（CM - 2）、一台将生成的几何数据渲染成图像的Cray Y - MP以及一台提供图像显示和用于指定几何生成及3D查看参数的用户界面的工作站组成。这三个元素通过使用几种不同的网络技术配置成一个虚拟系统。本文回顾了实现这个可配置环境所需的软件、硬件和通信技术的当前状况。这些相互依存的技术包括：（1）用户界面和应用程序构建方法；（2）用于连接应用程序软件模块的进程间通信（IPC）机制；（3）IPC用于在运行于单独且独立的计算系统元素上的模块之间进行通信的网络协议和接口硬件；（4）为连接应用程序所使用的地理上分布的元素的网络提供低级数据传输功能的电信基础设施；（5）将被连接以形成虚拟系统的功能元素的性质。
