Asian HPC Perspectives & Snapshot Dr. D.K. Kahaner Asian Technology Information Program (ATIP) email@example.com 25 June 2003 Outline • What is ATIP? • Key Trends – Illustrated in following slides – Earth Simulator – Special Purpose Computers in Japan – Grids everywhere – Biology as driver for new apps and markets • Further away: QC & biocomputing What is ATIP? • US not-for-profit company operating since 1994 • On-the-ground foreign S&T analysis (not only HPC) based on site visits, interviews, human networking, etc. • Reports, briefings, tours, seminars, liaison, ... • Offices / staff (Local multilingual scientists): Alb NM, JP, KR, TW, SG, IN, HK, AU, CN,... Lyon France [for ETIP] • Contractor to government agencies & companies • HPC major focus since inception • Reporting on HPC in Japan/Asia since 1989 • ATIP supports Asian participation in SC’XX • ATIP Sponsors HPC-Asia: TW, KR, SG, CN, AU, IN – HPC-Asia 2004 to be held in Japan – Steering committee members are key decision makers in public sector • WS in Beijing (9/03) “Programs & Plans for HPC Development in Asia” Overall Asian HPC Conclusions • High End: ES is brilliant accomplishment, new science will develop. US depth & breadth in HW, SW, apps, System Integration, will continue to provide leadership. • Real world: Significant capacity creation, SW/tools, System Integration, etc developing locally in most countries. HPC HW Development in Japan • Financed indirectly through government procurements (large supercomputer accounts in national research centers and universities) • Development of new large systems historically linked to major procurements at National Aerospace Laboratory (NAL), Japan Atomic Energy Research Institute (JAERI) • Future funding through government procurements is increasingly uncertain (university reform, tighter control of government R&D budget) • Also, since 1997, market for Japanese vendors is confined to Japan and Europe • Industry focus, as in the US is de-emphasizing processor development • Academic/research focus is on grids and bio apps ES General Comments • Custom vector processors & full crossbar have the potential to be very efficient on suitable problems, 65% system efficiency on large climate problem, 20% on nano problem. Fastest in the world (today). • Speed + efficiency leads to new science. • Credit to Dr Miyoshi for vision. • ES vs ASCI: ES is not a “program” but a single system. We don’t see much momentum for a follow-on. Nanotube Simulation Nakamura, et al Modeling Thermal Conductivity & Super Diamond Structures 3480PEs, 48,640 atoms, 5.47TF New Science • CNT properties – Thermal conductivity, depends on length of CNT, different from 3D materials – New material design – new nano-diamond, a light 3D semiconductor – Proof of stability of super diamond & fullerine encapsulated CNT (potential nano-memory) • Lower C60 melting point ~3400K, than other less detailed simulations • Resulted in funding for CNT-electronics simulation projects which are expected to lead to advances in electronic devices benefiting Japanese industry. Other Comments • From China: “The success of Japan’s Earth Simulator has shattered the US hegemon status in the supercomputer sector…. Consequently,… such architecture will gain renewed attention and its market share will possibly expand. Japan may keep its leading position on HPC performance in the coming two years and the US will readjust its supercomputer development scheme.” • Earth Simulator “buzz” is mostly from outside Japan. – However, publicity is certainly noted. • Japanese emphasis is on applications, not architecture. • Generated significant new interest in HPC in US, new programs, etc. – Positive impact on HPC community Special Purpose HW Under Development in Japan • Protein Explorer @ $ / Gflops W / Gflops Genomic Sciences Center Protein Explorer < 20 0.3 (2PF) BlueGene/L 140 6 – Fully completed in 2006 Pentium 4 PC 250 50 – Based on Grape, so is likely to be successful Earth Simulator 8000 128 • eHPC (“embedded” HPC) MDGRAPE-2 150 1.5 – Special purpose LSI – Computational science based on SOC (System-On- a-Chip) technology – Project developers mostly from “real” apps domains (comp chem, comp phys, nanotech, etc) Special Purpose HW Under Development in Japan Akira Suyama, University of Tokyo • Molecular and Biotechnological Computing • DNA computers, in particular, super parallel molecular computers using DNA molecular reactions are examined and constructed. • DNA computers are used for analyzing a class of problems known to be “difficult” for digital electronic computers, such as the NP perfect problem. • Additional projects include; a new molecule computer which utilizes a membrane structure, development of a logic circuit which operates autonomously and in parallel, and analysis of DNA genetic code. DNA Computer (Courtesy: A. Suyama.) • Protein Dynamics Simulation • The simulation of large flexible complex molecular systems, i.e., protein molecules are carried out by computer and the "dynamics" which mediates between structure and a function is studied. • Kinetic characteristics and functions, such as molecular motors are of special interest. Akira Suyama (cont.) • Gene Networks • The functions of specialization, aging, etc. analyzed by discovering the control network of these genes. • A new method for carrying out simultaneous measurement of the spatial / time change of a large number of genes, such as a DNA chip, is developed, and research which examines life phenomenon at the level of genes is performed. Olympus DNA Computer (Courtesy: Olympus • Genomic Information Science Optical Co., Ltd.) • Computer analysis using a genome database is performed including, searches for primitive genes, evolution of gene structure, and exploration of gene control networks. • Other Research • explored use of DNA computer for analysis in combination with DNA chips. • Developed “universal chip”, a DNA chip designed to indirectly measure designed sequences, e.g. DNA Coded Numbers, translated from raw information. • "intelligent DNA chip", which can perform logical reasoning and learning by using DNA computation on DNA Coded Numbers. JSPS Project on Molecular Computing (1996-2001) • US$60-80M per year from October 1996 to March 2001 • Part of Research for the Future Program -- themes of molecular computing, artificial cells, evolutionary computation, input/output for molecular computing and complex systems. • Theory of Molecular Computation: Theoretical results based on new computation paradigms such as splicing systems and self-organization. • Analysis of Molecular Computation and Its Design Strategy: Simulation, computational complexity, reaction mechanism, and sequence design of molecular computation. A new simulator, known as VNA (Virtual Nucleic Acid), was developed for reproducing molecular computation in a computer. • Molecular Implementation of Autonomous DNA Comp: Implemented automata with hairpin- formed DNA molecules, Computational model became known worldwide as Whiplash PCR. • Automated DNA Computation in Solid-Phase: Invented a solid-phase method which drastically reduces the number of DNA molecules required for molecular computations. The group constructed a DNA computer based on this technology. • Molecular Memory: Proposed the implementation of a write-once memory and its application, and named it `aqueous computing'. • Molecular Computer Based on Evolutionary Reactor: Molecular evolutionary reactor under the 3SR self-replication system w multiple enzyme to analyze evolution of DNA sequences. • Modeling Signal Pathways and Its Application to Molecular Computation: Computation using living cells, studied signal pathways and its modeling strategies. • DNA Nano-technology: Nishikawa, Grids and Clusters • Serious Cluster development – Planned: AIST SuperCluster (Tsukuba) • 10-29TF target – Existing: 108 cpu at TACC for QC Portal – Existing: 1064 cpu at CBRC • Grids are everywhere – Energetic discussions about their efficacy notwithstanding – Most countries (JP, CN, SG,TW, KR, AU) have national grid projects (some illustrations follow) GRIDS in Japan • SuperSINET • Information Technology Based Laboratory (ITBL) Project • Tokyo Institute or Technology Grid • Osaka Life Sciences Grid • RIKEN GSC Life Sciences Grid • 2003 Projects – Business Grid (funded by vendors + METI) aimed at supporting Japanese vendors “competing with IBM in Grid technology” (Kuwahara) – National Research Grid Initiative, NaReGI (funded by MEXT), HW for “theme” centers, incl nanotech, materials, life sciences – Grid Technology Research Center (GTRC) in Tsukuba (AIST): Goal: demonstrate feasibility of utility computing as a new approach to mainstream corporate computing • Will buy a large “supercluster” in 2003 • The recent GGF (Tokyo) is hosted by GTRC Bio in Japan follow the money... • Officially around US$3.7B, unofficially much more (proportional to US rel to pop) • Examples of New projects – MEXT: Link SNPs info & clinical data ($80M), DBs & SW for whole cell simulations ($40M) – METI: BioIT fusion, Structural Genomics, NanoBiotech, Structural analysis of glycoproteins ($65M) – ... • Many existing projects, e.g. – Brain Science ($80M), Protein 3000 (>$100M), Rice/insect genome... HPC Outside Japan • Looking to US for inspiration / systems / SW • No fundamental system development • Many examples of COTS system building – Legend (China) announces TF system (from Academia Sinica) – CDAC (India) PARAM 10000 TF system – Cluster purchase plan in HKBU • Many examples of basic research that may lead to future computing developments (plus, of course, QC such as at NTT) • HPC getting huge drive from biosciences in every Asian country China’s HPC Centers (2000) Beijing Xi’an Nanjing Chengdu Wuhan Shanghai Hefei Changsha Chinese HPC Companies (2002) • Top 3 domestic companies for low-end servers – Legend (spun off from ICT in 1981) – LangChao – Dawning (spun off from ICT in 1995) • Top 3 domestic companies for HPC servers – Dawning (entered HPC market in 1995) – Legend (entered HPC market in 2001) – LangChao (entered HPC market in 2002) • US vendors dominate but Chinese are developing indigenous capability • Many other HPC vendors now • Underpinnings: “China IT hardware production grows from $28.2B in 2001 to $35.2B in 2002, now No. 2 in the world” – No. 2 worldwide web audience – No. 2 PC sales Some Chinese Systems 2TFlops Legend Computer Shenwei-I, 384GFlops, Shanghai Acad of Math & System Sci, CAS, Beijing Supercomputer Center (512 P4 Xeon 2GHz CPU, Myrinet) (2D mesh distrib mem system) Grid Projects in China • 2002-2005 China National Grid Project supported by the 863 Plan of Ministry of Science & Technology (MOST) • The Chinese Academy of Sciences e-Science Grid • Vega Grid • The China Education Grid • Grid plans in Beijing and Shanghai • The “Next Internet” Project – Upgrade network infrastructure – Basic research in computing, data and access grids The China National Grid Project (2002-2005) • Dawning Grid Enabling Clusters (>4 TF) 4/03 • Grid Nodes (8-10 Nodes, Total 6-10 TF) • Grid Software (Grid OS, User Environment) • Grid Applications (resource integration and sharing) – Environment (Land Resources, Forestry, Pollution Control) – Industries (Aerospace, Automobile) – Service (Weather Forecasting) – Science (Bioinformatics, Drug Discovery, Basic Research) Dawning 3000 and Dawning 2000 Sunwei system Sun Microsystems machine SGI system China Market for HPC and Grid • Scientific Research – Bioinformatics, drug discovery, virtual observatory, traditional Chinese medicine, scientific database, basic sciences • Environment Protection and Natural Resources – Earth observation, forestry, pollution control, regional ecology • Manufacturing Industries – Aero, space, automobile, steel, regional SMEs • Services Sector – Railways, social security, labor markets, law enforcement, publications, communication, e-government, Digital Olympic • Common requirements: – Resource sharing and collaboration based on general-purpose, standards, open technology, not necessarily supercomputing China Remarks • China believes that HPC and Grid technology will reach mass adoption stage in less than 15 years • Two of the top markets will be China and India • Grid offers innovation opportunities • Note: China and ICT are keen to cooperate internationally, especially in A/P regions Singapore • Bio: SG has invested US$3.5B in biosci so far • SG Inst Mol Cell Bio managed by Sydney Brenner • Singapore Genomics Institute • SG Bioinformatics Institute Dir Gunarethnam Rajagopal – PhD Phys GaTech, Assist Dir Cambridge Cavendish Lab • SG Labs for Information Technology – GeneticXchange in Santa Clara to commercialize Klesli – BioMed Imaging Lab SG Institute for Bioengineering, SG BioProc Tech Center • SG’s Bio-Grid – Connect computing facilities & DBs at major life-sci res centers & hospitals – Long term goal to build large genome cohort repository to facilitate drug development. • Main problem for Singapore (general) is lack of qualified human resources Singapore, cont • iHPC (Inst of HPC), Nat Res Inst under A*STAR • Emphases: – Comp Mechanics (solid dynamics, CFD, multiphysics) – Comp Chem & Elec Systems (Materials, Electronics) – MEMS/NEMS modeling) – MINDEF collaborations • HW: – SGI Origin 2000 with 64 CPUs & 16 CPUs; – SUN E10000 with 40 CPUs & 29GB RAM; – IBM SP2 with 8 nodes & 10GB RAM; and – Linux Cluster with 30 (PIII 933Mhz, 1GB RAM) CPUs – 7 x IBM p690 (32CPUs) and total 512GB RAM – IBM Enterprise Storage System SAN with 4 TB hard-disk – CAVE • World-class center for engineering-modeling HPC apps Korea • Fastest (Dec 2002) cluster system @ Aerospace Structures Lab, Seoul National U, 180 Intel Xeons (1.6TF 64bit peak), 80th on the Nov 2002 Top 500. – Largely CFD, Comp Mech oriented • Nine computers in the Top 500 in Korea. India • Several parallel systems developed during the 1990s at independent institutes, all using COTs (from transputers to WSs) • Most fallen away. Center for Development of Advanced Computing (CDAC) remains, but has moved to developing system software and applications on commercial products HPC Facility @ C-DAC PARAM 10000 cluster of Sun Ultra e450 workstations Communication networks - Fast Ethernet, Myrinet and PARAMNet (in-house product PARAM Padma at Bangalore: 1TF Power 4 Processors (IBM) Storage System (Sun) CDAC’s System Area Network & CDAC’s System Software PARAM Padma • Compute Nodes based on the Power4 RISC processors, using Copper and SOI technology,in Symmetric Multiprocessor (SMP) configurations. • Nodes are connected through a primary high performance System Area Network, PARAMNet-II,designed and developed by C-DAC – Also a Gigabit Ethernet PARAM Padma • Runs C-DAC’s flexible and scalable HPCC software environment. • The Storage System designed to provide a primary storage of 5 Terabytes scalable to 22 Terabytes. • The network centric storage architecture, based on state-of-the-art Storage Area Network (SAN) technologies. • Uses Fibre Channel Arbitrated Loop (FC-AL) based technology for interconnecting storage subsystems like Parallel File Servers, NAS Servers, Metadata Servers, Raid Storage Arrays and Automated Tape Libraries, with I/O performance of up to 2Gigabytes/Second. • The Secondary backup storage subsystem is scalable from 10 to 100 Terabytes with automated tape library and support for DLT, SDLT and LTO Ultrium tape drivers. • Implements Hierarchical Storage Management (HSM) technology to optimize the demand on primary storage and effectively utilize the secondary storage • Accessible by users from remote locations. • PARAMNet-II Switch Low latency, high bandwidth interconnect, provides data rates of 2.5 Gigabits/sec in full duplex over fiber. The message latency is as low as 10 µsec. • Uses 16 port switch and a Network Interface Card (NIC) with an Application Programming Interface i.e. C-DAC’s Virtual Interface Provider Library(C- VIPL). • The non-blocking architecture of the switch allows multilevel switching for realizing a large cluster. The switch offers latency of the order of 0.5µsec. • Has 12 PARAMNet-II switches connected in two level configurations to form a 64-node CLOS network. PARAMNet-II Switch, cont. • NIC is based on C-DAC’s Communication Co-Processor-III (CCP- III) chip based on 0.15 micron 1M gates. • Implementation of packetization & reassembly, flow control, protection mechanism, address translation and error recovery in CCP-III, results in low latencies and very low overheads for the CPU. • Interfaces to SANSW8(8 port) and SANSW16 (16port) PARAMNet-II switch; Supports 2.5 Gbps(fibre)links; Host interface PCI 2.264bit/66MHz PARAM Padma CDAC Major Efforts • Parallelize commercial bio and engineering codes to run efficiently on PARAM 10000 and move to PARAM PADMA and develop suitable PSEs – Ex: CHARMM, AMBER, BLAST, Smith-Waterman, etc – Computational Atmospheric Sciences, Computational Fluid Dynamics, Computational Structural Mechanics,, Seismic Data Processing Bioinformatics, Quantum Chemistry ,Ab-initio Molecular Dynamics, Medical Imaging, Documentation Imaging, Machine Translation Server – Collaborate w Indian companies & univ – Ex: Strand Genomics: C.Delisi on Board RICCR • Russian Indian Center for Advanced Computing Research – CDAC ICAD (Moscow) Dir by Lenin Prize Winner (O.M. Belotserkovskii) – Aerohydrodynamics – Atmosphere and ocean – Weather forecasting – Seismic prospecting – Clean rooms – Forest Fires – Medicine Japan Semiconductor Special Projects That May Impact HPC Developments ASUKA 70 billion yen MIRAI • Total budget 30 billion yen - $800 million • From 2001 – 2007 HALCA 8 billion yen • Over 400 researchers METI’s Super Clean Room in Tsukuba (~$250M) Available for Japanese research community (nothing similar available in the US) New GaN High-frequency Device Development Project METI proposed (April 2003 – March 2004) Industrial Technology Budget Semiconductor Related Projects Next fiscal Current (US$ million dollars) year fiscal year budget budget (request) Next generation’s Semiconductor materials: Process – base project $66.0 $38.0 Semiconductor Application-chip project (New) – For High-end server $33.0 0 MRAM EUV System Project (New) $22.4 0 Super-advanced System LSI design $6.7 0 project What about software development? Apps & System tools mostly Apologies for too many slides in too little time... Thank you very much for your attention.
Pages to are hidden for
"Asian HPC Perspectives Snapshot Dr. D.K. Kahaner Asian"Please download to view full document