Los Alamos Computer Science Symposium Los Alamos Computer Science - PowerPoint PPT Presentation

LACSS08-1 Los Alamos Computer Science Symposium Los Alamos Computer Science Symposium (LACSS) (LACSS) � LANL is delighted to restart the Los Alamos Computer Science � LANL is delighted to restart the Los Alamos Computer Science Symposium series after a 1 year interruption (following the passing of Symposium series after a 1 year interruption (following the passing of Prof. Ken Kennedy who chaired the previous Symposia) Prof. Ken Kennedy who chaired the previous Symposia) � Motivation for the Symposium: � Motivation for the Symposium: � � serve as a premier international event showcasing research and serve as a premier international event showcasing research and development in high-performance computing development in high-performance computing � � foster collaborations between national labs, academia and industry in the foster collaborations between national labs, academia and industry in the HPC arena HPC arena � � promote computer science research that will advance HPC programs at promote computer science research that will advance HPC programs at LANL and elsewhere LANL and elsewhere � � encourage a strong focus on HPC in the academic computer science encourage a strong focus on HPC in the academic computer science community community � Symposium theme this year : hybrid and heterogeneous systems, � Symposium theme this year : hybrid and heterogeneous systems, including architecture and programming environments . including architecture and programming environments . Operated by the Los Alamos National Security, LLC for the DOE/NNSA Unclassified

LACSS08-2 Programs Stewardship Capabilities Operations Capabilities Operations Global Security Emerging Threats FY 2008 LANL Budget: $2.2 Billion FY 2008 LANL Budget: $2.2 Billion Operated by the Los Alamos National Security, LLC for the DOE/NNSA Unclassified

LACSS08-3 LANL’s future is multi-program, based on LANL’s future is multi-program, based on integrated capabilities integrated capabilities Be the premier National Security Science Laboratory for a dynamic future. • We must succeed in three broad areas of growth and transformation How We Work Transform our Science that Matters How We Work Transform our Science that Matters Scientific Campus Scientific Campus � Collaborate, partner and � Collaborate, partner and � Information Science & � Information Science & and Facilities and Facilities team to make decisive team to make decisive Technology enabling Technology enabling � Campus for 2020 � Campus for 2020 contributions to our contributions to our Integrative and Integrative and sponsors sponsors Predictive Science � Consistent with Predictive Science � Consistent with complex complex � Outstanding operational � Outstanding operational � Experimental Science � Experimental Science transformation transformation excellence for safety, excellence for safety, focused on Materials for focused on Materials for security, and efficient security, and efficient � LANSCE-R, CMRR, � LANSCE-R, CMRR, the Future the Future pursuit of ST&E for our pursuit of ST&E for our Science Complex Science Complex � Fundamental Forensic missions � Fundamental Forensic missions � Signature Facilities: � Signature Facilities: Science for Nuclear, Science for Nuclear, MaRIE and Roadrunner MaRIE and Roadrunner Biological, Chemical Biological, Chemical Threats Threats Operated by the Los Alamos National Security, LLC for the DOE/NNSA Unclassified

LACSS08-4 Information Science and Technology at LANL Integration for a data-rich accelerated discovery and predictive capability Operated by the Los Alamos National Security, LLC for the DOE/NNSA Unclassified

LACSS08-5 Connecting the Dots in Science Connecting the Dots in Science Finding the Dots Understanding the Dots Providing Predictive Sheer Volume of Data Advanced Mathematics Climate Understanding and Algorithms Now: 20-40 Terabytes/year � Produce hydrogen-based energy � Huge dimensional space 5 years: 5-10 Petabytes/year � Stabilize carbon dioxide � Combinatorial challenge Fusion � Clean and dispose toxic waste � Complicated by noisy data Now: 100 Megabytes/15 min � Requires high-performance 5 years: 1000 Megabytes/2 min computers Raymond L. Orbach, DOE Undersecretary for Science 2006 AAAS Annual Meeting Operated by the Los Alamos National Security, LLC for the DOE/NNSA Unclassified

LACSS08-6 LANL History in HPC Hardware and Advanced LANL History in HPC Hardware and Advanced Architectures for HPC–Enabled S&T Frontiers Architectures for HPC–Enabled S&T Frontiers Robust scalable computing and Thinking Machines - We started communications the modern Era PaScalBB Mult-Lane Environment Archive throughput scalable with fail over 20-40 GBytes/sec at least, with fail 1985 Lane 1 Subnet Archive Core sized over KRB switches, Lane 2 Subnet VLANS,virtual Lane 3 Subnet DNS1 routers, for each N-way ether trunk Lane 4 Subnet color, OSPF or 10G NFS I/O node To Site OSPF, need 2 Network minimum per Single file Lane system Global scalable file system, scale >> 1 SW Cluster A, interconnect Use DHCP- RELAY for broadcast functions Cluster A needs more bandwidth per lane than other Layer 2-3 switches clusters Lightning - Modern Linux Clusters Cluster C, interconnect Cluster B, interconnect Heterogeneous Computing Roadrunner - to Petaflops 2006 Operated by the Los Alamos National Security, LLC for the DOE/NNSA Unclassified

LACSS08-7 LANL has been exploring hybrid computing, a LANL has been exploring hybrid computing, a transformational technology, for over 5 years transformational technology, for over 5 years � Roadrunner embodies many � Roadrunner embodies many key architectural trends key architectural trends � Cluster computing � Cluster computing � Multicore � Multicore � SIMD � SIMD � Hybrid � Hybrid � Roadrunner provides � Roadrunner provides immediate benefits for immediate benefits for solving important problems solving important problems in weapons science in weapons science Roadrunner is a different path to a petascale system Operated by the Los Alamos National Security, LLC for the DOE/NNSA Unclassified

LACSS08-8 Petascale performance provided by Roadrunner will Petascale performance provided by Roadrunner will push scientific boundaries in numerous areas. push scientific boundaries in numerous areas. astrophysics ocean modeling turbulence Roadrunner Final system fluid instabilities protein magnetic dynamics reconnection Operated by the Los Alamos National Security, LLC for the DOE/NNSA Unclassified

LACSS08-9 Roadrunner embodies many key architectural Roadrunner embodies many key architectural trends, each in moderation trends, each in moderation � Roadrunner has: multicore, short-vector SIMD, threads, � Roadrunner has: multicore, short-vector SIMD, threads, heterogeneous instruction sets, local stores instead of heterogeneous instruction sets, local stores instead of caches, on-chip CPU/memory networks, remote caches, on-chip CPU/memory networks, remote accelerators and cluster computing. accelerators and cluster computing. � Any of these features can be used as needed, without � Any of these features can be used as needed, without requiring extremes in any one of them. requiring extremes in any one of them. � Roadrunner ’s scale and flexibility makes it an ideal base � Roadrunner ’s scale and flexibility makes it an ideal base from which to explore the changing landscape of hybrid from which to explore the changing landscape of hybrid and heterogeneous architectures. and heterogeneous architectures. Operated by the Los Alamos National Security, LLC for the DOE/NNSA Unclassified

LACSS08-10 A LANL Update for LACSS A LANL Update for LACSS To emphasize the importance of this area for LANL endeavors, To emphasize the importance of this area for LANL endeavors, we have established the Center for Advanced Architectures we have established the Center for Advanced Architectures and Usable Supercomputing (CAAUS) and Usable Supercomputing (CAAUS) � LACSS is organized under the auspices of CAAUS � LACSS is organized under the auspices of CAAUS � Some goals of CAAUS: � Some goals of CAAUS: � comprehensive focus on the design of new generations of high-productivity � comprehensive focus on the design of new generations of high-productivity supercomputing systems grounded in applications of interest to LANL for the supercomputing systems grounded in applications of interest to LANL for the 21st century 21st century � build a world-class computer science capability for HPC at Los Alamos � build a world-class computer science capability for HPC at Los Alamos � attract staff at Los Alamos in computer science, and serve as a portal for � attract staff at Los Alamos in computer science, and serve as a portal for staff integration into other technical activities at the laboratory staff integration into other technical activities at the laboratory � build new programs to support LANL goals in HPC R&D � build new programs to support LANL goals in HPC R&D Operated by the Los Alamos National Security, LLC for the DOE/NNSA Unclassified