A Distributed Computing Facility

该基础设施奖用于分布式计算设施的建设。该网络由桌面工作站组成，这些工作站通过中速网络连接到高速网络上的后端计算和数据存储服务器。后端资源由中速计算服务器、共享内存并行多处理器和大规模并行机器组成。该基础架构支持的研究包括使用Nuprl支持的构造型理论在应用逻辑方面的工作；在科学计算中，开发有效利用分布式和大规模并行计算资源的算法；在机器人应用的建模和仿真；以及分布式计算软件。实验计算机科学中的许多问题需要单个工作站无法提供的峰值资源。这些资源可以是计算周期或内存。这个问题的一个新兴解决方案是利用工作站和计算服务器网络中的空闲处理器。这个解决方案的实现仍然需要对将问题分解成小块的机制进行大量研究，以便在分配计算以充分利用可用处理器的同时最小化通信开销。康奈尔大学的分布式计算研究人员将使用他们的分布式计算设施作为分布式算法的测试平台。康奈尔大学的三个活跃研究领域将特别受益于这一设施。应用逻辑小组使用康奈尔大学开发的一个名为Nuprl的系统来支持建设性推理。通过使用这个工具，组合学和编程语言中的开放性问题已经得到了解答。科学计算小组已经参与了一个名为LAPACK的软件包的构建，该软件包用于线性代数例程，涉及为并行和分布式机器架构构建新的算法。最后，建模和仿真组需要大量的计算资源，使他们能够将计算资源用作实验测试平台，从而在不花费建造机器人的费用的情况下测试设计。

This infrastructure award is for the construction of a distributed computing facility. The network consists of desktop workstations connected over a medium speed network to back-end computation and data-storage servers on a high-speed network. The back-end resources consist of mid-speed compute servers, shared memory parallel multiprocessors, and massively parallel machines. The research supported by this infrastructure includes work in applied logic using the constructive type theory supported by Nuprl; in scientific computing in developing algorithms that effectively utilize distributed and massively parallel computing resources; in modeling and simulation for robotics applications; and in software for distributed computing. Many problems in experimental computer science require peak resources not available on a single workstation. These resources could be computation cycles or memory. An emerging solution to this problem is to utilize idle processors in a network of workstations and compute servers. The realization of this solution still requires much research into mechanisms for breaking problems into pieces so as to minimize communication overhead while distributing the computation to make best use of the available processors. The distributed computing researchers at Cornell University will use their distributed computing facility as a testbed for distributed algorithms. Three active areas of research at Cornell will especially benefit from this facility. The applied logic group uses a system called Nuprl developed at Cornell to support constructive reasoning. With the use of this tool, open questions in combinatorics and programming language have already been answered. The scientific computing group is already involved in the construction of a software package called LAPACK for linear algebra routines that involves constructing new algorithms for parallel and distributed machine architectures. Finally the modeling and simulation group requires substantial computational resources to enable them to use the computing resources as an experimental testbed allowing designs to be tested without the expense of constructing a robot.