CISE Research Instrumentation: Infrastructure for Cluster and Parallel Systems Research

EIA-9986020 Rudolf EigenmannPurdue UniversityCISE Research Instrumentation：Infrastructure for Cluster and Parallel Systems Research普渡大学电气与计算机工程学院将收购四台Sun Ultra Enterprise 450四处理器服务器以及Myrinet互连，用于计算机和信息科学与工程研究。 这些设备将用于几个研究项目，特别是： 分布式共享存储器的推测性分布式共享存储器延迟减少的并行技术，对称多处理器集群的大规模应用研究复用：未来的十亿晶体管微体系结构新设备将为细粒度软件分布式共享存储器（DSM）提供一个测试平台，以实现和评估推测性一致DSM体系结构，并作为编译器的目标。 此外，集群和单个服务器也将在单用户模式下使用，对我们的大规模应用程序进行无干扰的测量，并允许单用户访问单个SMP，而不会抑制长时间运行的多用户应用程序。服务器将提供所需的内存和处理能力以及内存带宽，以执行Multiplex架构的内存和计算机密集型模拟。新设备将通过同时运行多个模拟来提供高模拟吞吐率。集群和单个服务器将提供高内存带宽、大聚合内存和处理能力，以运行详细的、大规模的并行模拟，用于推测性DSM研究。

EIA-9986020Rudolf EigenmannPurdue UniversityCISE Research Instrumentation: Infrastructure for Cluster and Parallel Systems ResearchThe School of Electrical and Computer Engineering at Purdue University will acquire four Sun Ultra Enterprise 450 quad-processor servers plus Myrinet interconnections for research in computer and information science and engineering. The equipment will be used for several research projects, including in particular: Speculative Distributed Shared Memory Latency-Reducing Compiler Techniques for Distributed Shared Memory, Studies of large-scale Applications for Symmetric Multiprocessor Clusters Multiplex: Future Billion-Transistor MicroarchitecturesThe new equipment will provide a testbed platform for fine-grain software distributed shared memory (DSM) to implement and evaluate speculative coherent DSM architectures and serve as a target for a compiler. In addition, the cluster and individual servers will also be used in single-user mode to perform unperturbed measurements of our large-scale applications and will allow single-user access to individual SMPs without inhibiting long-running multi-user applications.The servers will provide the required memory and processing power as well as memory bandwidth to perform the memory-and computer-intensive simulations of the Multiplex architecture. The new equipment will deliver high simulation throughput rates by running many simulations concurrently. The cluster and individual servers will provide the high memory bandwidth, large aggregate memory and processing power required to run detailed, large-scale parallel simulations for the speculative DSM research .