SCIENCE
Appro Wins Huge Deal with Three National Laboratories for 438 Teraflops Clusters
Tri-Laboratory Linux Capacity Cluster (TLCC07) program joins effort for next generation Linux clusters featuring easily replicated four-socket system scalability: Appro, a leading provider of high-performance enterprise computing systems, today announced the historic award of Appro Xtreme-X high performance computing clusters for the TLCC07 project to be delivered to three National Nuclear Security Administration (NNSA) weapons laboratories: Lawrence Livermore National Laboratory, Los Alamos National Laboratory and Sandia National Laboratories. This represents the first time, under a single contract, that the tri-Labs purchase the same high performance computing systems for deployment at all three sites. 
The TLCC07 program represents a multi-million dollar contract between Lawrence Livermore National Laboratory and Appro to provide scalable Linux high performance clusters with up to 438 teraFLOP/s in 21 Scalable Units (SU) with an option for an additional 10 SU, 200 teraFLOP/s of capacity computing power to be deployed in 8 separate Linux clusters at the tri-Lab sites. This project consists of a total of 3,024 nodes, 12,096 processors/48,348 cores with up to 96.8TB of memory. These Appro high performance clusters will be based on Quad-Core AMD Opteron processors connected with a two stage InfiniBand 20 Gb/s 4X Double Data Rate (DDR) fabric featuring Voltaire Grid Director 288 port core switches. These clusters feature the latest Mellanox Technologies ConnectX IB 20Gb/s dual-port InfiniBand adapters and ConnectX EN dual port 10 Gigabit Ethernet NICs for storage cluster connectivity. These high performance clusters will be deployed to the tri-Labs community starting in late 2007 through early 2008. These scalable Linux clusters will be used principally to provide needed computational support to NNSA’s nuclear weapons programs, notably Stockpile Stewardship – the program to ensure the safety, security and reliability of the nation’s nuclear deterrent without nuclear testing. Additional capacity computing is needed for such efforts as completion of the National Ignition Facility laser fusion project by 2010; implementation of updated computer codes for nuclear weapons certification; and the program to extend the life of existing weapons in the stockpile. "This is a win-win for Appro and the DOE," said Steve Conway, IDC research vice president, technical computing. "With its modular Scalable Unit architecture, Appro captured a large, 12,000-plus processor, highly sought-after HPC procurement and will assist the DOE in its critical core mission. The DOE employed a comprehensive and extensive purchasing strategy for their capacity computing needs, with a focus on exploiting the economies of scale and taking the longer view of total cost of ownership. IDC expects that cluster-based supercomputing will soon represent over 75% of all technical servers." The TLCC07 program started when Advanced Simulation and Computing Program Office in collaboration with Lawrence Livermore, Los Alamos, and Sandia National Laboratories introduced the project to significantly reduce Total Cost of Ownership (TCO) for deploying several large-scale production capacity compute clusters at the Tri-Lab sites. The key to this effort is standardizing the hardware and software for next generation Linux clusters at the Tri-Labs. On the hardware side, this standardization is based on a new innovative Scalable Units (SU) design that allows rapid manufacture, delivery, acceptance and integration of multiple Linux clusters. The idea is that by purchasing a total of 21 SUs (437 teraFLOP/s) over two years, tremendous economies of scale are generated to reduce TCO components of SU cost, integration time and manpower, hardware, system and application software support costs. This huge deal with the Laboratories represents a third contract that Appro was awarded over the last two years. In aggregate with these cluster deliveries, Appro will provide to the National Labs over 13 Linux clusters of 543 TeraFLOP/s of capacity computing power with a total of 72,452 processor cores and over 143.9 TB of memory. These new Appro high performance computing clusters will be deployed based on the innovative Scalable Units design and will be configured and installed with the same architecture as Appro deployed in the previous awarded contracts. Appro plans to incorporate this Scalable Supercomputing Cluster Architecture and launch the Appro Xtreme-X Supercomputer series at the Supercomputing show scheduled for Nov 12 – 15th in Reno, NV 2007. “The ability to provide a scalable unit in multiple scalable clusters to all three national Lab sites with a 30-50% lower TCO is of great value to NNSA’s effort to apply high performance computing to time-urgent national security challenges,” said Mark Seager, Advanced Computing Technology lead for LLNL. “In the past, high-end Linux clusters were more expensive and difficult to integrate and deploy. By defining this Scalable Unit “building block” approach to structure these multiple cluster deployments using COTS parts, the tri-Labs are able to significantly reduce TCO, simplify the integration and deployment of multiple Linux clusters at all three Laboratories and put highly needed capacity clusters into production on an accelerated timeline.” "These three National Laboratories have a long, rich history of fielding the world's most powerful supercomputers to tackle some of the world's most difficult problems, said Daniel Kim, CEO of Appro International. By combining common hardware and software technologies at all three laboratories using Appro’s supercomputing clusters, we are pleased to meet the demands of petascale computing and beyond while providing the reliability required by the NNSA/ASC Program Office with price/performance leadership.” “Quad-Core AMD Opteron processors offer outstanding scalability, especially in four socket systems. It also delivers the memory bandwidth performance that is critical to HPC applications, based on the benefits of Direct Connect Architecture – an integrated memory controller combined with HyperTransport technology,” said David Rich, director of High Performance Computing, AMD. “The new features in our enhanced micro-architecture, such as a dedicated floating point unit per core, in concert with the flexibility and scalability of Appro’s system design, can help these national labs reach world-class results in computational output.” “We are very pleased to work with Appro again to deliver a world-class supercomputer to the tri-Labs that will advance their mission-critical national security computing efforts,” said Ronnie Kenneth, Chairman and CEO, Voltaire. “Voltaire’s 20 Gb/second InfiniBand solutions increase the bandwidth and messaging rate of the system even at the petascale level so the Laboratories’ applications can run faster and more efficiently with the reliability they require.” “Mellanox’s 20Gb/s ConnectX IB InfiniBand adapters ideally match the Tri-Lab’s requirements for high-performance clustering I/O with industry-leading throughput and low-latency, scalability, and efficiency. We’re also pleased that our performance-leading ConnectX EN 10 Gigabit Ethernet adapters have been selected to provide additional storage I/O connectivity for the cluster,” said Eyal Waldman, chairman, president, and CEO of Mellanox Technologies. “This win highlights Mellanox’s continual growth as the leading interconnect for large-scale cluster deployments.”
The TLCC07 program represents a multi-million dollar contract between Lawrence Livermore National Laboratory and Appro to provide scalable Linux high performance clusters with up to 438 teraFLOP/s in 21 Scalable Units (SU) with an option for an additional 10 SU, 200 teraFLOP/s of capacity computing power to be deployed in 8 separate Linux clusters at the tri-Lab sites. This project consists of a total of 3,024 nodes, 12,096 processors/48,348 cores with up to 96.8TB of memory. These Appro high performance clusters will be based on Quad-Core AMD Opteron processors connected with a two stage InfiniBand 20 Gb/s 4X Double Data Rate (DDR) fabric featuring Voltaire Grid Director 288 port core switches. These clusters feature the latest Mellanox Technologies ConnectX IB 20Gb/s dual-port InfiniBand adapters and ConnectX EN dual port 10 Gigabit Ethernet NICs for storage cluster connectivity. These high performance clusters will be deployed to the tri-Labs community starting in late 2007 through early 2008. These scalable Linux clusters will be used principally to provide needed computational support to NNSA’s nuclear weapons programs, notably Stockpile Stewardship – the program to ensure the safety, security and reliability of the nation’s nuclear deterrent without nuclear testing. Additional capacity computing is needed for such efforts as completion of the National Ignition Facility laser fusion project by 2010; implementation of updated computer codes for nuclear weapons certification; and the program to extend the life of existing weapons in the stockpile. "This is a win-win for Appro and the DOE," said Steve Conway, IDC research vice president, technical computing. "With its modular Scalable Unit architecture, Appro captured a large, 12,000-plus processor, highly sought-after HPC procurement and will assist the DOE in its critical core mission. The DOE employed a comprehensive and extensive purchasing strategy for their capacity computing needs, with a focus on exploiting the economies of scale and taking the longer view of total cost of ownership. IDC expects that cluster-based supercomputing will soon represent over 75% of all technical servers." The TLCC07 program started when Advanced Simulation and Computing Program Office in collaboration with Lawrence Livermore, Los Alamos, and Sandia National Laboratories introduced the project to significantly reduce Total Cost of Ownership (TCO) for deploying several large-scale production capacity compute clusters at the Tri-Lab sites. The key to this effort is standardizing the hardware and software for next generation Linux clusters at the Tri-Labs. On the hardware side, this standardization is based on a new innovative Scalable Units (SU) design that allows rapid manufacture, delivery, acceptance and integration of multiple Linux clusters. The idea is that by purchasing a total of 21 SUs (437 teraFLOP/s) over two years, tremendous economies of scale are generated to reduce TCO components of SU cost, integration time and manpower, hardware, system and application software support costs. This huge deal with the Laboratories represents a third contract that Appro was awarded over the last two years. In aggregate with these cluster deliveries, Appro will provide to the National Labs over 13 Linux clusters of 543 TeraFLOP/s of capacity computing power with a total of 72,452 processor cores and over 143.9 TB of memory. These new Appro high performance computing clusters will be deployed based on the innovative Scalable Units design and will be configured and installed with the same architecture as Appro deployed in the previous awarded contracts. Appro plans to incorporate this Scalable Supercomputing Cluster Architecture and launch the Appro Xtreme-X Supercomputer series at the Supercomputing show scheduled for Nov 12 – 15th in Reno, NV 2007. “The ability to provide a scalable unit in multiple scalable clusters to all three national Lab sites with a 30-50% lower TCO is of great value to NNSA’s effort to apply high performance computing to time-urgent national security challenges,” said Mark Seager, Advanced Computing Technology lead for LLNL. “In the past, high-end Linux clusters were more expensive and difficult to integrate and deploy. By defining this Scalable Unit “building block” approach to structure these multiple cluster deployments using COTS parts, the tri-Labs are able to significantly reduce TCO, simplify the integration and deployment of multiple Linux clusters at all three Laboratories and put highly needed capacity clusters into production on an accelerated timeline.” "These three National Laboratories have a long, rich history of fielding the world's most powerful supercomputers to tackle some of the world's most difficult problems, said Daniel Kim, CEO of Appro International. By combining common hardware and software technologies at all three laboratories using Appro’s supercomputing clusters, we are pleased to meet the demands of petascale computing and beyond while providing the reliability required by the NNSA/ASC Program Office with price/performance leadership.” “Quad-Core AMD Opteron processors offer outstanding scalability, especially in four socket systems. It also delivers the memory bandwidth performance that is critical to HPC applications, based on the benefits of Direct Connect Architecture – an integrated memory controller combined with HyperTransport technology,” said David Rich, director of High Performance Computing, AMD. “The new features in our enhanced micro-architecture, such as a dedicated floating point unit per core, in concert with the flexibility and scalability of Appro’s system design, can help these national labs reach world-class results in computational output.” “We are very pleased to work with Appro again to deliver a world-class supercomputer to the tri-Labs that will advance their mission-critical national security computing efforts,” said Ronnie Kenneth, Chairman and CEO, Voltaire. “Voltaire’s 20 Gb/second InfiniBand solutions increase the bandwidth and messaging rate of the system even at the petascale level so the Laboratories’ applications can run faster and more efficiently with the reliability they require.” “Mellanox’s 20Gb/s ConnectX IB InfiniBand adapters ideally match the Tri-Lab’s requirements for high-performance clustering I/O with industry-leading throughput and low-latency, scalability, and efficiency. We’re also pleased that our performance-leading ConnectX EN 10 Gigabit Ethernet adapters have been selected to provide additional storage I/O connectivity for the cluster,” said Eyal Waldman, chairman, president, and CEO of Mellanox Technologies. “This win highlights Mellanox’s continual growth as the leading interconnect for large-scale cluster deployments.” 