The Extensible Terascale Facility (ETF), or TeraGrid involves nine resource providers operating high-performance computing, information, and visualization facilities, interconnected with a dedicated high bandwidth optical network and integrated with a set of common software that presents the user community with a cohesive set of distributed resources and services. On October 1, 2004, the ETF will conclude a three-year construction effort and will transition into an operational state. This proposal outlines services to be provided at NCSA as part of early operation of the TeraGrid system.
NCSA brings a tremendous wealth of experience in production HPC environments to the ETF partnership. Not only will NCSA be providing the largest scale high-end systems to TeraGrid, but also experienced personnel who will provide the users with support throughout the life of their projects. NCSA staff process initial allocations requests, route and respond to questions and problem reports, assist scientists with their computational and data needs, tune the TeraGrid environment for optimal performance, and promote science success stories. Additionally, through the ETF construction project, NCSA has provided considerable contributions to ETF in the areas of file system and system scaling, file transfer performance characterization, as well as contributed to building the core TeraGrid security team
NCSA will be providing an aggregate 16 TFlops of peak computing power and over half a petabyte of online storage for ETF. NCSA will initially provide a 10TFlop IBM/Intel Itanium 2 compute resource connected to one parallel file system on a 120TB Storage Area Network (SAN) system, and a second parallel file system across the entire cluster attached to servers with 60TB of direct-connected online storage. A second production computing resource, a shared-memory, symmetric multi-processor (SMP) SGI Altix system, will come online in March 2005. This system will consist of 1,024 Intel Itanium 2 processors running the Linux operating system, 3 terabytes of globally accessible memory, and 370 terabytes of SGI Infinite Storage, with a peak performance of more than 6 teraflops.
