SCI: Terascale Computing System

匹兹堡超级计算机中心将获得，提出，整合，稳定，评估和测试一组具有代表性的初始应用的早期模型（原型）超级计算系统。该系统是由克雷设计和建造的，是他们的红色风暴原型模型。这个原型模型被设计成一个高度可扩展的能力系统，非常苛刻，高端计算科学和工程研究应用。它将与PSC文件和网络系统连接并提供服务，并将成为NSF TeraGrid网络基础设施项目的一个组成部分。这个系统足够大，可以支持广泛的测试，以建立其可行性，在更大的规模上，作为美国研究界的领导级资源。PSC“红色风暴”将是克雷公司为桑迪亚国家实验室（SNL）建造的“红色风暴”系统的改进版本。PSC“红色风暴”系统将包括大约2000个计算节点，峰值性能约为每秒10万亿次浮点运算。每个节点包括一个先进的64位AMD Opteron处理器和一个专有的网络通信ASIC。节点被组织成一个非常高速的三维环形信息传递网络。整个系统设计包括一个精心设计的并行RAS（可靠性、可用性、可服务性）功能，这将确保非常高的可靠性和可管理性。系统软件套件包括用于计算节点的轻量级内核、用于服务和IO节点的Linux以及完整的编程环境。在硬件和软件方面，本系统尽可能采用COTS组件。这个项目不仅包括对一个特别有前途的架构的适用性的调查，还包括应用研究工作和有用的专题研究的表现。该项目包括基础设施（设备和仪器）网络、合作伙伴（评估和测试小组），如果成功，将通过升级为额外的基础设施奠定基础。PSC讲习班将提供宣传和额外培训。

The Pittsburgh Supercomputer Center will acquire, bring up, integrate, stabilize, assess and test on a representative set of initial applications an early model (prototype) supercomputing system This system is designed and built by Cray and is their Red Storm prototype model. This prototype model is designed to be a highly scalable capability system of very demanding, high end computational scientific and engineering research applications. It will be interfaced to, and served by, the PSC file and network systems and will be a component of the NSF TeraGrid Cyberinfrastructure project. This system is sufficiently large to support a broad range of testing to establish its viability, at a larger scale, as a leadership-class resource for the U. S. Research Community. The PSC Red Storm will be a modified version of the Red Storm system now under construction by Cray, Inc. for the Sandia National Laboratory (SNL).The PSC Red Storm system will comprise approximately 2,000 computational nodes for a peak performance of approximately ten teraflops. Each node includes an advanced 64-bit AMD Opteron processor and a proprietary network communications ASIC. The nodes are organized in a very high-speed, 3-D toroidal message-passing network. The overall system design includes an elaborate parallel RAS (Reliability, Availability, Serviceability) capability which will ensure very high reliability and manageability. The system software suite includes a light-weight kernel for compute nodes, Linux for the service and IO nodes, and a full programming environment. Both in hardware and software, this system uses COTS components where feasible. This project includes not only the investigation of the suitability of a particularly promising architecture but also involves application research work and the performance of useful topical research. The project includes infrastructure (facilities and instrumentation) networks, partnerships (evaluation and testing groups) and, if successful, lays the groundwork for additional infrastructure through upgrades. PSC workshops will provide dissemination and additional training.