Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Nikravesh big datafeb2013bt

HPC and Big Data 2013

  • Be the first to comment

  • Be the first to like this

Nikravesh big datafeb2013bt

  1. 1. The Emergence of Computation for Interdisciplinary Large Data Frontiers in HPC and Data Analytics inspired by Science Bounded by our imagination innovation through Technology Create Social impact Masoud Nikravesh @ LBNL and Maxeler Mnikravesh@lbl.gov Nikravesh@Maxeler.com Visiting Scientist- Lawrence Berkeley National Lab Vice President- Maxeler Technologies Health, Freshwater, Food Security, Ecosystems, and Urban Metabolism 1
  2. 2. Outline of Talk Drivers for Change: Computing and Big Data Computational Science and Engineering Big Data and State Economic Model The State New Economy Model State-wide Initiative Maxeler Technologies 2
  3. 3. Outline of Talk Drivers for Change: Computing and Big Data Computational Science and Engineering Big Data and Economic Model The State New Economy Model State-wide Initiative Maxeler Technologies 3
  4. 4. Drivers for Change- Computation • Continued exponential increase in computational power  simulation (Computing) is becoming third pillar of science, complementing theory (Analytic and Math ) and experiment (Applications) Applications HPC-Cloud Computing Analytics Math High performance computing (HPC), large-scale simulations, and scientific applications all play a central role in CSE. CSE The HPC/cloud computing initiative and next generation data center Extreme simulation, visual-data analytics, data-enabled scientific discovery Applications/real‐world complex applications (scientific, engineering, social, economic, policy) using the future multi-core parallel computing ((i.e. E-Informatics, Earthquake Early Warning, NextGenMaps, Genome Atlas, Genetic Facebook, Genomics Browser) HPC-Petascale and Exascale systems are an indispensable tool for exploring the frontiers of science and technology for social impact. 4
  5. 5. Moore’s Law is Alive and Well 2X transistors/Chip Every 1.5 years Called “Moore’s Law” Moore’s Law Microprocessors have become smaller, denser, and more powerful. Gordon Moore (co-founder of Intel) predicted in 1965 that the transistor density of semiconductor chips would double roughly every 18 months. Slide source: Jack Dongarra 5
  6. 6. But Clock Scaling Bonanza Has Ended Processor designers forced to go “multicore”:  Heat density: faster clock means hotter chips  more cores with lower clock rates burn less power  Declining benefits of “hidden” Instruction Level Parallelism (ILP)  Last generation of single core chips probably over-engineered  Lots of logic/power to find ILP parallelism, but it wasn’t in the apps  Yield problems  Parallelism can also be used for redundancy  IBM Cell processor has 8 small cores; a blade system with all 8 sells for $20K, whereas a PS3 is about $600 and only uses 7 6
  7. 7. Clock Scaling Hits Power Density Wall 4004 8008 8080 8085 8086 286 386 486 Pentium® P6 1 10 100 1000 10000 1970 1980 1990 2000 2010 Year PowerDensity(W/cm2) Hot Plate Nuclear Reactor Rocket Nozzle Sun’s Surface Source: Patrick Gelsinger, Intel Scaling clock speed (business as usual) will not work 7
  8. 8. Revolution is Happening Now  Chip density is continuing increase ~2x every 2 years  Clock speed is not  Number of processor cores may double instead  There is little or no more hidden parallelism (ILP) to be found  Parallelism must be exposed to and managed by software Source: Intel, Microsoft (Sutter) and Stanford (Olukotun, Hammond) 8
  9. 9. Computing Growth is Not Just an HPC Problem 10 100 1,000 10,000 100,000 1,000,000 1985 1990 1995 2000 2005 2010 2015 2020 Year of Introduction The Expectation Gap Microprocessor Performance “Expectation Gap” over Time (1985-2020 projected) 9
  10. 10. New Processors Means New Software  Exascale will have chips with thousands of tiny processor cores, and a few large ones  Architecture is an open question:  sea of embedded cores with heavyweight “service” nodes  Lightweight cores are accelerators to CPUs  Autotuning eases code generation for new architectures Interconnect Memory Processors Server Processors Manycore processors 130 Megawatts 75 Megawatts Source: Kathy Yelick, 10
  11. 11. New Processor Designs are Needed to Save Energy  Server processors have been designed for performance, not energy  Graphics processors are 10-100x more efficient  Embedded processors are 100-1000x (1.25 rather than 100 watt)  Need manycore chips with thousands of cores Cell phone processor (0.1 Watt, 4 Gflop/s) Server processor (100 Watts, 50 Gflop/s) Source: Kathy Yelick, HPC-SEG July 2011 11
  12. 12. Source: Oliver Pell, HPC-SEG July 2011, Berkeley CPU, GPU, Hybrid, FPGA? 12
  13. 13. x86 Multicores GPU FPGA Numbers -Current generation: 4–6 cores/CPU x 2 CPUs/node = 8–12 cores/node -Future generation: 16–20 cores/CPU x 4 CPUs/node = 64–80 cores/node -512 cores/GPU (Nvidia) -1600 cores/GPU (AMD) -No more cores but BRAM, --Look Up Tables, FlipFlops, etc.. -Clock frequency is in the order of hundreds of MHz -Memory per card is in the order of tens of GB What is the easy part? -Well known and mature technology -Well established development environments -Parallelism between core and nodes -Well known technology (for gaming purposes) -It is becoming reliable also for HPC computation -High performance-per-watt ratio What is difficult to do? -Linear speedup with increasing core numbers -CUDA: good tool but proprietary -OpenCL: open technology but not yet standard and more complex to use -Development tools (+ profiling, debugging, etc) not yet fully available -Non standard development tools (VHDL is not for Geophysicists… but we have MaxCompiler!) -Data streaming technology is different from standard approaches (grid/matrix) Main problems -Slow memory access -Legacy codes need to be re-engineered in order to get the best performance (e.g. SSE vectorization, cache blocking) -Network connections have to be optimized for the architecture -Limited amount of memory (4–6 GB) per card -Slow communication with the host CPU (due to PCI Express) -Internal bandwidth is not always enough -The technology is not yet standard for HPC -Slow communication with the host CPU (due to PCI Express) Source: Carlo Tomas, HPC-SEG, July 2011, Berkeley 13
  14. 14. A Likely Trajectory - Collision or Convergence? CPU GPU multi-threading multi-core many-core fixed function partially programmable fully programmable future processor by 2012 ? programmability parallelism after Justin Rattner, Intel, ISC 2008 14
  15. 15. Interconnect Memory Processors New Memory and Network Technology to Lower Energy  Memory as important as processors in energy  Latency is physics, bandwidth is money  Software managed memory or cache hybrids  Autotuning has helped with that management  Need to raise level of autotuning to higher level kernels Usual memory + network New memory + network 25 Megawatts75 Megawatts Source: Kathy Yelick, 15
  16. 16. goal usual scaling 2005 2010 2015 2020 Energy Cost Challenge for Computing Facilities At ~$1M per MW, energy costs are substantial  1 petaflop in 2010 will use 3 MW  1 exaflop in 2018 possible in 200 MW with “usual” scaling  1 exaflop in 2018 at 20 MW is DOE target 16
  17. 17. Exascale: Who Needs It? Fusion: Simulations of plasma properties to ITER scale model Combustion: complete predictive engine simulation Astronomy: origins of the universe Sequestration: Understanding fluid flow & chemistry Materials: solar panels to database of materials-by-design. Climate: Resolve clouds (1km scale) & model mitigations Protein structures: From Biofuels to Alzheimers Every field needs more computing! 1) To quantify and reduce uncertainty in simulations 2) Analyze data from experiments and simulations 17
  18. 18. TOP10 Sites – Nov 2012 Rank Site System Cores Rmax (TFlop/s) Rpeak (TFlop/s) Power (kW) 1 DOE/SC/Oak Ridge National Laboratory United States Titan - Cray XK7 , Opteron 6274 16C 2.200GHz, Cray Gemini interconnect, NVIDIA K20x Cray Inc. 560640 17590.0 27112.5 8209 2 DOE/NNSA/LLNL United States Sequoia - BlueGene/Q, Power BQC 16C 1.60 GHz, Custom IBM 1572864 16324.8 20132.7 7890 3 RIKEN Advanced Institute for Computational Science (AICS) Japan K computer, SPARC64 VIIIfx 2.0GHz, Tofu interconnect Fujitsu 705024 10510.0 11280.4 12660 4 DOE/SC/Argonne National Laboratory United States Mira - BlueGene/Q, Power BQC 16C 1.60GHz, Custom IBM 786432 8162.4 10066.3 3945 5 Forschungszentrum Juelich (FZJ) Germany JUQUEEN - BlueGene/Q, Power BQC 16C 1.600GHz, Custom Interconnect IBM 393216 4141.2 5033.2 1970 6 Leibniz Rechenzentrum Germany SuperMUC - iDataPlex DX360M4, Xeon E5-2680 8C 2.70GHz, Infiniband FDR IBM 147456 2897.0 3185.1 3423 7 Texas Advanced Computing Center/Univ. of Texas United States Stampede - PowerEdge C8220, Xeon E5-2680 8C 2.700GHz, Infiniband FDR, Intel Xeon Phi Dell 204900 2660.3 3959.0 8 National Supercomputing Center in Tianjin China Tianhe-1A - NUDT YH MPP, Xeon X5670 6C 2.93 GHz, NVIDIA 2050 NUDT 186368 2566.0 4701.0 4040 9 CINECA Italy Fermi - BlueGene/Q, Power BQC 16C 1.60GHz, Custom IBM 163840 1725.5 2097.2 822 10 IBM Development Engineering United States DARPA Trial Subset - Power 775, POWER7 8C 3.836GHz, Custom Interconnect IBM 63360 1515.0 1944.4 3576 18
  19. 19. 19
  20. 20. 20
  21. 21. TOP500 Sites – June 2011 Today, HPC-Petascale and soon Exascale systems- is not just a tool of choice, but it becomes an indispensable tool for frontiers of science and technology for social impact. Petaflop with ~1M Cores in your PC by 2025? 8-10 years 6-8 years 21
  22. 22. Drivers for Change – Big Data • Continued exponential increase in experimental, simulation, sensors, and social data  techniques and technology in data analysis, visualization, analytics, networking, and collaboration tools are becoming essential in all data rich applications Big Data Model Human Experts- Citizen Cyber Science Crowdsourceing Analytic ToolsFirst Principles Hybrid Models Google IBM-Watson IBM- Cognitive Model Boeing 747 Simulation Protein Folding Amazon AI-ImageIncreased climate/environmentaldetail Increased socio-economic detail Tera Peta Peta Exa Socio-Economic Modeling for Large-scale Quantitative Climate/Environmental Change Analysis En Informatics Environment-Genetic 22
  23. 23. World Population: Today-~6B, 2050-~9B, 2100-~10B %70 will live in Cities by 2050 By 2020: 35 trillion Gigabytes Data (Cyber-Physical world is connected through billions to even trillions of sensors and devices) Petaflop with ~1M Cores in your PC by 2025? Health, Freshwater, Food Security, Ecosystems, and Urban Metabolism 23
  24. 24. Why BIG Data is a Big Deal? Size of Data: • 2010: 1.2 million Petabytes, or 1.2 Zettabytes • 2020: 35 trillion Gigabytes (Cyber-Physical World is connected through billions to even trillions of sensors and devices) Type of data: • from homogenous data to heterogeneous and multi-scale • from physical sensor data to social-economical data • from complete to incomplete, imprecise and uncertain • from implementing on single-simple hardware-software architecture to scalable parallel complex hardware-software architectures 24
  25. 25. Why BIG Data is a Big Deal? Crisis: Data storage/transfer/communication and security- privacy doomsday forecast Opportunities: Information gold mine Needs: better, faster, cheaper, and scalable technologies for storage, manipulation, communication and analysis 25
  26. 26. Why BIG Data is a Big Deal? Challenge: Combine our current and to be developed advanced-scalable* analytical tools with first principle models and human capabilities at scale with anticipatory capabilities to discover the un-seen phenomena and insights and to make and deliver securely right decisions and at the right time based on incomplete, imprecision, and uncertain public/private data dealing with multi and conflicting objectives and criteria. 26
  27. 27. Why BIG Data is a Big Deal? Crowdsourcing Big Data Model Human Experts- Citizen Cyber Science Crowdsourceing Analytic ToolsFirst Principles Hybrid Models Google IBM-Watson IBM- Cognitive Model Boeing 747 Simulation Protein Folding Amazon AI-Image Increased climate/environmentaldetail Increased socio-economic detail Tera Peta Peta Exa Socio-Economic Modeling for Large-scale Quantitative Climate/Environmental Change Analysis En Informatics Environment-Genetic 27
  28. 28. Distributed thinking / Human computing Physical participation coordinated via Internet BIG Data and Citizen Cyber Science? What can be aggregated?  Aggregate perception, knowledge, reasoning  Visual pattern recognition  Real-world knowledge  3D spatial manipulation  Language skills Where to get Volunteers  Tell a good story about your research  Give recognition  Make it a game  Add a social dimension 28
  29. 29. Cloud Computing Cloud Computing are being used by a broad array of Computational Science and Engineering faculty investigators, researchers and graduate students from social scientists and economists to astrophysicist and Bioengineers. 29
  30. 30. What is a Cloud? Definition-NIST 30 According to the National Institute of Standards & Technology (NIST)… Resource pooling. Computing resources are pooled to serve multiple consumers. Broad network access. Capabilities are available over the network. Measured Service. Resource usage is monitored and reported for transparency. Rapid elasticity. Capabilities can be rapidly scaled out and in (pay-as-you-go) On-demand self-service. Consumers can provision capabilities automatically.
  31. 31. What is a cloud? Cloud Models 31
  32. 32. Map Reduce  Map:  Accepts input key/value pair  Emits intermediate key/value pair  Reduce :  Accepts intermediate key/value* pair  Emits output key/value pair Very big data Result M A P R E D U C E Partitioning Function
  33. 33. Workflow
  34. 34. Partitioning Function
  35. 35. Parallelism
  36. 36. MapReduce: The Map Step vk k v k v map vk vk … k v map Input key-value pairs Intermediate key-value pairs … k v
  37. 37. MapReduce: The Reduce Step k v … k v k v k v Intermediate key-value pairs group reduce reduce k v k v k v … k v … k v k v v v v Key-value groups Output key-value pairs
  38. 38. Distributed Execution User Program Worker Worker Master Worker Worker Worker fork fork fork assign map assign reduce read local write remote read, sort Output File 0 Output File 1 write Split 0 Split 1 Split 2 Input Data
  39. 39. Cloud Infrastructure Applications (scientific, engineering, social, economic/business/finance, policy) Delivery of Services Mobile Devices Mobile CloudSoftware and Appliances Cluster Scheduling & Reliability Network Research and Security Supercomputer Public Cloud Private Cloud Volunteering Computing Mobile Cloud Streaming Data Massive Data Extreme Simulation Large Scale Visualization Machine Learning Analytics Intelligent Dynamic Maps Early Warning Social Networking Second Life Cyber Citizen Personalized Services Crowd Sourcing Cloud Computing 39
  40. 40. Cloud Computing  Infrastructure – Cloud Cluster and Data Centers  Delivery of Services – Mobile Cloud  Applications  Scientific  Social  Economics/Business  Software and Appliances  Cluster Scheduling & Reliability  Network Research and Security Mobile devices, Mobile Cloud, and Cloud Infrastructure will be the device/tools of choice for delivery of services. 40
  41. 41. Cloud Computing Initiative The focus will be on three main areas:  Machine Learning: Provide the general public with machine learning analytics tools and algorithm runs in cloud infrastructure.  Streaming Data Analytics and Visualization: Analyses and visualization of large-scale real time data sets such as traffic information, online news sources, economics data, and scientific data such as astrophysical and Genomics data.  Scientific Applications: Benchmarking and cataloging the suitability of cloud computing for science and engineering applications, including HPC applications. 41
  42. 42. BIG Data and Sensors/Cyber-Physical Infrastructure Water Air Energy Earthquake Marvell Lab μSensors TinyOS Prototyping Devices and Sensors G/H FEEDBACK California Independent System (Cal ISO) Department of Water Resources California Department of Health and Social Services and FCC Cyberspace Handhelds Laptop/PC Clusters IBM/ room143 Cloud + + + Analytics Algorithms M/C Learning/A.I. Statistical Analysis Social Comp Knowledge Insight Large-Scale Information Extraction Delivery and Service Back to Handhelds Distributed Systems Visualization, Analytics and Insight Physical World Big Data Streams Nano Lab Clusters 42
  43. 43. Increased climate/environmentaldetail Increased socio-economic detail Tera Peta Peta Exa Socio-Economic Modeling for Large-scale Quantitative Climate/Environmental Change Analysis En Informatics Environment-Genetic BIG Data and Exa-Scale Computing 43
  44. 44. Courtesy of U.S. Department of Energy Human Genome Program , http://www.ornl.gov/hgmis BIG Data and DNA Computing 44
  45. 45. BIG Data and DNA Computing 45
  46. 46. BIG Data and DNA Computing 46
  47. 47. BIG Data and Visualization –Scientific 47
  48. 48. BIG Data and Visualization - Business 48
  49. 49. Outline of Talk Drivers for Change: Computing and Big Data Computational Science and Engineering Big Data and Economic Model The State New Economy Model State-wide Initiative Maxeler Technologies 49
  50. 50. Computational Science Nature, March 23, 2006 “An important development in sciences is occurring at the intersection of computer science and the sciences that has the potential to have a profound impact on science. It is a leap from the application of computing … to the integration of computer science concepts, tools, and theorems into the very fabric of science.” -Science 2020 Report, March 2006 50
  51. 51. Computational Science and Engineering 51
  52. 52. What is CSE? CSE is a rapidly growing multidisciplinary field that encompasses real-world complex applications (scientific, engineering, social, economic, policy), computational mathematics, and computer science and engineering. High performance computing (HPC), large-scale simulations and modeling (physical, biological, economic, social, and policy processes), and scientific applications all play a central role in CSE. Petaflop with ~1M Cores in your PC by 2025? 52
  53. 53. What is CSE? Simulation of complex problems is sometimes the only feasible way to make progress if the theory is intractable and experiments are too difficult, too expensive, too dangerous, or too slow. Through modeling and simulation of multiscale systems of systems, and through scientific discovery from large-scale heterogeneous data, CSE aims to advance solutions for a wide range of problems in the areas of nanoscience and nanotechnology, energy, climate change, engineering design, neuroscience, cognitive computing and intelligent systems, plasma physics, transportation, bioinformatics and computational biology, earthquake engineering, geophysical modeling, astrophysics, materials science, national defense, information technology for health care, engineering better search engines, socio-economic-policy modeling, and other fields that are critical to scientific, economic, and social progress. 53
  54. 54. CSE: Vision To support the work of scientists and engineers as they pursue complex –simulation/modeling, as well as computational, data and visualization- intensive research to enhance scientific, technological, and economic leadership while improving our quality of life. inspired by Science Bounded by our imagination innovation through Technology Create Social impact Today, HPC-Petascale and soon Exascale systems- is not just a tool of choice, but it becomes an indispensable tool for frontiers of science and technology for social impact. 54
  55. 55. CSE: Mission  Conduct world-leading research in applied mathematics and computer science to provide leadership in such areas as energy, environment, health- information technology, climate, bioscience and neuroscience, and intelligent cyber-physical infrastructure to name a few.  Be at the forefront of the development and use of ultra-efficient largest-scale computer systems, focusing on discoveries and solutions that link to the evolution of the commercial market for high-performance and cloud computing and services.  Allow industry collaborators to gain experience with computational modeling / simulation and the effective use of HPC and Cloud facilities and carrying back new expertise to their institutions. This would enable the Industry partners to be “first to market” with important scientific and technological capabilities, breakthrough ideas, and new hardware-software.  Educate the next generation of interdisciplinary students and industry leaders (DE-CSE program and a new Professional Master Program (PMS) to be developed) inspired by Science Bounded by our imagination innovation through Technology Create Social impact Petaflop with ~1M Cores in your PC by 2025? 55
  56. 56. High performance computing (HPC), large-scale simulations, and scientific applications all play a central role in CSE. Applications HPC-Cloud Computing Analytics Math CSE The HPC/cloud computing initiative and next generation data center Extreme simulation, visual-data analytics, data-enabled scientific discovery Applications/real‐world complex applications (scientific, engineering, social, economic, policy) using the future multi-core parallel computing ((i.e. E-Informatics, Earthquake Early Warning, NextGenMaps, Genome Atlas, Genetic Facebook, Genomics Browser) CSE HPC-Petascale and Exascale systems are an indispensable tool for exploring the frontiers of science and technology for social impact. 56
  57. 57. Nature of Work, Education and Future Society “Creative Creators” or “Creative Servers”: Do complex task, and Enhance, Refine, and Reinvent. “T. Friedman and M. Mandelbaum” That Used to be Us” 20th Century 21th Century Number of Jobs 1-2 Jobs 10-15 Jobs Job Requirement Mastery of one Field (Single Deep Expertise) Breadth; Depth in several Fields (Multiple Deep Expertise) (Broad Knowledge) Alternative sources of Natural Resources: Energy and Water Technology: Nano-technology, Quantum Computers, Genetic and Biometrics, and Robotics Services: Online Education and Services on Demand Resources: Sensors and Devices, Big Data, Computing Power, Social Network and Computing Charles Fadel 57
  58. 58. Tm T m Tm-shaped Individual and not just T or m-shaped Single Expertise Multiple Deep Expertise Single Deep + Multiple Expertise Hybrid (CSE) Broad Knowledge 21st century skills: problem-solving, critical thinking, entrepreneurship and creativity 58
  59. 59. Educating the Workforce of the Future China & India: 300M Skilled worker by 2025 Eng. Ph.D Median Salary: India: $39,200 China: $53,700 Germany: $99,400 US(CA): $125,200 Science and Engineering Graduate US 420000, EU 470000, China 530000 , India 690000, Japan 350000 McKinsey report concluded that only 10% of Chinese engineers and 25% of Indian engineers can compete in the global outsourcing arena. Revised by: Nikarvesh 59
  60. 60. Annualized Job Openings vs. Annual Degrees Granted (2008-2018) CSE educates the next generation of interdisciplinary students and industry leaders. CSE Revised by: Nikarvesh 60
  61. 61. Degree Production vs. Job Openings Sources: Adapted from a presentation by John Sargent, Senior Policy Analyst, Department of Commerce, at the CRA Computing Research Summit, February 23, 2004. Original sources listed as National Science Foundation/Division of Science Resources Statistics; degree data from Department of Education/National Center for Education Statistics: Integrated Postsecondary Education Data System Completions Survey; and NSF/SRS; Survey of Earned Doctorates; and Projected Annual Average Job Openings derived from Department of Commerce (Office of Technology Policy) analysis of Bureau of Labor Statistics 2002-2012 projections. See http://www.cra.org/govaffairs/content.php?cid=22. 160,000 140,000 120,000 100,000 80,000 60,000 40,000 20,000 Engineering Physical Sciences Biological Sciences Computer Science Ph.D. Master’s Bachelor’s Projected job openings CSE educates the next generation of interdisciplinary students and industry leaders. CSE Revised by: Nikarvesh 61
  62. 62. Open Big Data Science Computational Foundations and Driving Applications Open Big Data Science APPS CORE LIBRARIES ANALYTICS MACHINE LEARNING TRANINING & EDUCATION OUTREACH Devices and Computing Environment 62
  63. 63. Center will develop a wide array of computational tools to tackle the challenges of data-intensive scientific research across multiple scientific disciplines. These tools will encapsulate state of the art machine learning and statistical modeling algorithms into broadly applicable, high-level interfaces that can be easily used by application scientists. The goal is to dramatically reduce the time needed to extract knowledge from the floods of data science is facing, thanks to workflows that permit exploratory and collaborative research to evolve into robustly reproducible outcomes. Data-Driven Scientific Computing 63
  64. 64. The development will be driven by a collection of scientific problems that share a common theme. They all present major data-intensive challenges requiring significant algorithmic breakthroughs and represent key questions within their field, from rapid astronomical discovery of rare events to early warning systems for natural hazards such as earthquakes or tsunamis. Moving beyond the traditional domain of scientific computing, we will tackle a collection of problems in social sciences and the digital humanities, pushing the boundaries of quantitative scholarship in these disciplines. Center for Data-Driven Scientific Computing 64
  65. 65. Accelerating Environmental Synthesis and Solutions (ACCESS) & Environment Quality and Security To enable synthesis, En Informatics (En= Environmental, Ecological, Epidemiological, Economic, Engineering, Equitable, Ethical,… ) Health, Freshwater, Food Security, Ecosystems, and Urban Metabolism World Population: Today-~6B, 2050-~9B, 2100-~10B %70 will live in Cities by 2050 65
  66. 66. ACCESS Focus ACCESS will focus on five major domains critical for human welfare and environmental quality: freshwater, health, ecosystems, urban metabolism, and food security; and will create and implement a synthesis process that makes research tools and understanding rapidly accessible across disciplines, and foster new ways of thinking across disciplines about critical environmental problems. Accelerating Environmental Synthesis and Solutions (ACCESS) 66
  67. 67. ACCESS Themes Ecosystem trajectories over the past million years and in the future - rate and nature - result principally 8000 generations of human population growth and aspirations. Underlying ecosystem trajectories are the changing supply and demand of water and the need to harness energy to advance civilization. Urban metabolism: Theoretical models of cities as complex socio- ecological systems with particular metabolic dynamics. Urban policy is increasingly critical to building a more sustainable future. The increasing ease of utilizing existing resources leads to their rapid and unsustainable depletion, with many resulting intolerable impacts, including those on  Human and animal health  Food security Center for Accelerating Environmental Synthesis and Solutions (ACCESS) 67
  68. 68. Urban Metabolism Conceptual Frameworks for Urban Metabolism: Theoretical models of cities as complex socio-ecological systems with particular metabolic dynamics include approaches based in political economy, sociology, urban ecology and biogeochemistry, and industrial ecology – many of which remain disconnected from each other. In addition, because the inputs to urban life are globalized, the geography of consumption and production networks must be integrated into conceptual frameworks. Data Integration: A rapidly expanding volume of geospatial data on urban stocks and flows – about people, animals, vegetation, consumer products, energy, waste, etc. – is available for synthesis and building models of the complex metabolic cycles of cities. Policy and Activism: Urban policy is increasingly critical to building a more sustainable future, but the policy interventions and activist campaigns are piecemeal remedies rather than solutions based on an understanding of cities as complex socio-ecological systems. Visualization and Decision-Support: Decision makers and stakeholders of many types need to visuzlize model results quickly and effectively. Generating sophisticated and insightful visualizations of urban systems is an emergent and critical field. 68
  69. 69. Insight Lab Applications Machine Learning Massive Scale Data Analytics and Visualization Data Structure Analytics Service Delivery 69
  70. 70. Strategic Projects/ Shared Facilities, Resources, Expertise Technology Streaming Data and Visual Analytics Core Group* Core Scientific Group* Shared Facilities VisLab+ Computing Infrastructures Delivery of Service Mobile Devices, Internet, and Cloud Science/Applications scientific,engineering,social,economic/business/finance ACCESS- E-informatics Earthquake Early Warning Next Generation Dynamic Maps Genome Atlas, Genetic Facebook, Genomics Browser, bioinformatics, Immune System, … Computational Bioscience, Neuroscience, Nanoscience , Astrophysics , … *core group of enabling computational scientists would stand at the heart of the center, and that they would both cross- pollinate expertise among projects and provide great leverage in winning large federally-supported projects*. Educational, Research, and Social Impacts; IT-Enabled Disaster Resilience Insight Lab Intensive Computing, Immersive Visualization and Human Interaction Data and Visual-enabled Scientific Discovery and Insight Accelerator 70
  71. 71. Outline of Talk Drivers for Change: Computing and Big Data Computational Science and Engineering Big Data and Economic Model The State New Economy Model State-wide Initiative Maxeler Technologies 71
  72. 72. List of U.S. States by Unemployment Rate State or District Unemployment rate (seasonally adjusted) Monthly percent change (=drop in unemployment) Nevada 12.6 0.4% California 11.1 0.2% Rhode Island 10.8 0.3% Mississippi 10.4 0.1% District of Columbia 10.4 0.2% North Carolina 9.9 0.1% Florida 9.9 0.1% Illinois 9.8 0.2% Georgia 9.7 0.1% South Carolina 9.5 0.4% Michigan 9.3 0.5% Kentucky 9.1 0.3% Indiana 9.0 0.0% New Jersey 9.0 0.1% Oregon 8.9 0.2% Arizona 8.7 0.0% Tennessee 8.7 0.4% Washington 8.5 0.2% Idaho 8.4 0.1% United States (mean)[5] 8.3 0.2% Connecticut 8.2 0.2% Alabama 8.1 0.6% Ohio 8.1 0.4% New York 8.0 0.0% Missouri 8.0 0.2% Colorado 7.9 0.1% West Virginia 7.9 0.0% State or District Unemployment rate (seasonally adjusted) Monthly percent change (=drop in unemployment) United States (mean)[5] 8.3 0.2% Texas 7.8 0.3% Arkansas 7.7 0.2% Pennsylvania 7.6 0.3% Delaware 7.4 0.2% Alaska 7.3 0.0% Wisconsin 7.1 0.2% Maine 7.0 0.0% Massachusetts 6.8 0.2% Louisiana 6.8 0.1% Montana 6.8 0.3% Maryland 6.7 0.2% New Mexico 6.6 0.1% Hawaii 6.6 0.1% Kansas 6.3 0.2% Virginia 6.2 0.0% Oklahoma 6.1 0.0% Utah 6.0 0.4% Wyoming 5.8 0.0% Minnesota 5.7 0.2% Iowa 5.6 0.1% Vermont 5.1 0.2% New Hampshire 5.1 0.1% South Dakota 4.2 0.1% Nebraska 4.1 0.0% North Dakota 3.3 0.1% January 24, 2012 for December 2011 Source: Wikipedia 72
  73. 73. The State New Economy Index* Methodology The State New Economy Index uses 26 indicators. These Indicators are divided into five categories. These categories best capture what is new about the New Economy: 1) Knowledge Jobs (5) 2) Globalization (2) 3) Economic Dynamism (3.5) 4) Transformation to a Digital Economy (3) 5) Technological Innovation Capacity (5) *Source: ITIF-Kauffman 73
  74. 74. Top 10 US States ranked based on “The New Economy Index” 2010 1. Massachusetts (92.6) 2. Washington (77.5) 3. Maryland (76.9) 4. New Jersey (76.9) 5. Connecticut(76.6) 6. Delaware (75.0) 7. California (74.3) 8. Virginia (73.7) 9. Colorado (72.8) 10. New York (71.3) 2008 1. Massachusetts (97) 2. Washington (81.9) 3. Maryland (80) 4. Delaware (79.3) 5. New Jersey (77) 6. Connecticut (76.1) 7. Virginia (75.6) 8. California (75) 9. New York (74.4) 10. Colorado (70.4) 2007 1. Massachusetts (96.1) 2. New Jersey (86.4) 3. Maryland (85.0) 4. Washington (84.6) 5. California (82.9) 6. Connecticut (81.8) 7. Delaware (79.6) 8. Virginia (79.5) 9. Colorado (78.3) 10. New York (77.4) 2002 1. Massachusetts (90.0) 2. Washington (86.2) 3. California (85.5) 4. Colorado (84.3) 5. Maryland (75.6) 6. New Jersey (75.1) 7. Connecticut (74.2) 8. Virginia (72.1) 9. Delaware (70.5) 10. New York (69.3) 1999 1. Massachusetts (82.3) 2. California (74.3) 3. Colorado (72.3) 4. Washington (69.0) 5. Connecticut (64.9) 6. Utah (64.0) 7. New Hampshire (62.5) 8. New Jersey (60.9) 9. Delaware (59.9) 10. Arizona (59.2) 74
  75. 75. ITIF-Kauffman Ranking 26 Attributes PCA (MNIK2012) 5 Categories PCA (MNIK2012) Massachusetts Massachusetts Massachusetts Washington Washington New Jersey Maryland Connecticut Connecticut New Jersey Maryland Washington Connecticut New Jersey Maryland Delaware Virginia Delaware California California California Virginia Colorado Virginia Colorado Delaware New York New York New Hampshire Colorado New Hampshire Minnesota New Hampshire Utah Utah Minnesota Minnesota New York Utah Oregon Oregon Oregon Illinois Illinois Illinois Rhode Island Michigan Rhode Island Michigan Rhode Island Texas Texas Pennsylvania Michigan Georgia Texas Georgia Arizona Vermont Florida Florida Arizona Pennsylvania Pennsylvania Georgia Arizona Vermont North Carolina Vermont North Carolina Ohio North Carolina ITIF-Kauffman Ranking 26 Attributes PCA (MNIK2012) 5 Categories PCA (MNIK2012) Ohio Idaho Kansas Kansas Kansas Ohio Idaho Wisconsin Nevada Maine Florida Maine Wisconsin Missouri Idaho Nevada Nebraska Wisconsin Alaska New Mexico Alaska New Mexico Maine Missouri Missouri Iowa Nebraska Nebraska Alaska Hawaii Indiana North Dakota Indiana Montana Hawaii Iowa North Dakota Indiana North Dakota Iowa South Carolina New Mexico South Carolina Nevada Tennessee Hawaii South Dakota South Carolina Tennessee Tennessee Montana Oklahoma Montana Louisiana Kentucky Oklahoma Oklahoma Louisiana Wyoming Kentucky South Dakota Alabama South Dakota Wyoming Kentucky Wyoming Alabama Louisiana Alabama Arkansas Arkansas Arkansas West Virginia West Virginia West Virginia Mississippi Mississippi Mississippi US States ranked based on “The New Economy Index” and two new PCA ranking models!?? 75
  76. 76. KNOWLEDGE JOBS Weight IT Professionals Professional and Managerial Jobs Workforce Education Immigration of Knowledge Workers U.S. Migration of Knowledge Workers Manufacturing Value-Added Traded-Services Employment GLOBALIZATION Export Focus on Manufacturing and Services Foreign Direct Investment (FDI) ECONOMIC DYNAMISM Job Churning Initial Public Offerings (IPOs) Entrepreneurial Activity Inventor Patents Fastest-Growing Firms The State New Economy Index* DIGITAL ECONOMY Online Population Digital Government Farms and Technology Broadband Health IT INNOVATION CAPACITY High-Tech Employment Scientists and Engineers Patents Industry R&D Non-industry R&D Green Economy Venture Capital Ref.*: ITIF and Kauffman Foundation 76
  77. 77. Knowledge Job (5) 1 Massachusetts (17.39) 2 Connecticut (16.78) 3 Maryland (15.40) 4 Virginia (15.37) 5 Delaware (13.94) 6 Minnesota (13.94) 7 New Jersey (13.85) 8 Washington (13.80) 9 New York (13.66) 10 New Hampshire (12.96) 13 California (10.70) Top 10 US States ranked based on “The New Economy Index” Globalization (2) 1 Delaware (18.05) 2 Texas (16.39) 3 South Carolina (15.31) 4 New Jersey (14.73) 5 Connecticut (14.68) 6 Massachusetts (14.59) 7 Kentucky (14.24) 8 New York (14.21) 9 Washington (13.73) 10 North Carolina (13.61) 17 California (13.17) Economic Dynamism (3.5) 1 Utah (14.94) 2 Colorado (13.74) 3 Georgia (13.38) 4 Massachusetts (13.30) 5 Florida (13.09) 6 Montana (12.87) 7 Arizona (12.64) 8 Nevada (12.56) 9 California (12.01) 10 Idaho (11.86) Digital Economy (3) 1 Massachusetts (16.40) 2 Rhode Island (15.53) 3 New Jersey (15.13) 4 Maryland (14.29) 5 Connecticut (14.09) 6 California (14.07) 7 New York (14.03) 8 Oregon (13.58) 9 Washington (13.41) 10 Virginia (12.82) Innovation Capacity (5) 1 Massachusetts (19.0) 2 Washington (17.5) 3 California (15.0) 4 Maryland (13.4) 5 Delaware (13.1) 6 Colorado (13.0) 7 New Hampshire (12.2) 8 New Jersey (12.2) 9 Virginia (12.0) 10 New Mexico (11.8) 77
  78. 78. Projection of the cases on the factor-plane ( 1 x 3) Cases with sum of cosine square >= 0.00 Active AL AK AZ AR CA CO CT DE FL GA HI ID IL IN IAKS KY LA ME MD MA MI MN MS MO MT NE NV NH NJ NM NY NC ND OH OK OR PA RI SC SD TN TX US UT VT VA WA WV WI WY -8 -6 -4 -2 0 2 4 6 Factor 1: 34.46% -2 -1 0 1 2 3 4 Factor3:10.00% AL AK AZ AR CA CO CT DE FL GA HI ID IL IN IAKS KY LA ME MD MA MI MN MS MO MT NE NV NH NJ NM NY NC ND OH OK OR PA RI SC SD TN TX US UT VT VA WA WV WI WY Top 25 States Bottom 25 States PCA Analysis of US States Ranking: The New Economy Index (26 Indicators) 78
  79. 79. Outline of Talk Drivers for Change: Computing and Big Data Computational Science and Engineering Big Data and Economic Model The State New Economy Model State-wide Initiative Maxeler Technologies 79
  80. 80. State-Wide Initiative  building upon massive scale datasets – streaming and static (sensors/social-economic)  employing sophisticated analytics, with an emphasis on modeling, simulation, and crowdsourcing  focus on major domains critical for human welfare and environmental quality (Environment and Security); urban metabolism and smart cities, food security, fresh water resources, public health, natural disasters, energy conservation, and ecosystem.  educating the next generation of interdisciplinary students and industry leaders A statewide initiative to create integrated systems and advanced analytic tools using advanced computational science and engineering 80
  81. 81. States can improve the standard of living by applying predictive simulation systems and integrated advanced analytic tools using advanced computational science and engineering to critical problems facing the states How can States respond to rapidly changing environment, climate change, socio-economic forces and demographics?  water resources, public health, natural disasters, energy conservation, environment and security Predictive simulation and advanced analytic can be used to  understand the impacts of policy choices  understand social and economical impacts  create new technologies and industries  find more efficient solutions to California’s pressing infrastructure problems Health, Freshwater, Food, Energy, Environment Security, Ecosystems, and Urban Metabolism 81
  82. 82. Outline of Talk Drivers for Change: Computing and Big Data Computational Science and Engineering Big Data and Economic Model The State New Economy Model State-wide Initiative Maxeler Technologies 82
  83. 83. Maxeler Technologies 83
  84. 84. Roots at Stanford University, Bell Labs, and Imperial College London Founded in 2003, incorporated in Delaware and England 2006: signs long term R&D contract with Chevron in San Ramon CA 2010: ENI (Italy) buys largest Maxeler Supercomputer for Imaging 2011: sold 20% stake to JP Morgan’s strategic investments group Maxeler Technologies 2012: Partnership for Sequence Assembly and Analysis with EU Genomics Center 84
  85. 85.  Oskar Mencer, CEO, previously at Technion, Stanford, DIGITAL, Hitachi and Bell Labs.  Prof Michael J Flynn, Chairman, Professor Emeritus, Stanford, previously VC partner, Founder of American Supercomputers, and manager at IBM.  Stephen Weston, Chief Development Officer, previously Managing Director at JP Morgan, Deutsche Bank, UBS, and CS.  Over 50 employees, over 10 PhDs and scientists  Main office is in London, UK.  see www.maxeler.com Maxeler Technologies – The Team 85
  86. 86. 86 Maxeler Technologies - Divisions
  87. 87. The Challenges- HPC We are approaching the end of easy scaling as predicted by Moore’s Law, reaching limits in multiple dimensions: Power, Space, Time, and Cost. The Power Gap: As CPU transistors shrink, the net effect is an increase in density but consequently power consumption is on the rise too The Space Gap: The space required to perform computation continues to expand as our appetite for solving complex problems marches on The Time Gap: As we explore new science and exploit Big Data, we also increase application complexity and as a direct result runtime The Cost Gap: Each server node added to the data center increases operational costs in the form of utility rates 87
  88. 88. The Maxeler Solution: How We Deliver Maximum Performance Computing Maxeler has bridged these gaps to deliver maximum performance by designing systems to meet the needs of the application rather than forcing applications to conform to a generic machine This approach optimises for performance whilst minimising on space, cost, and power Power Gap  Time Gap Space Gap  Cost Gap  88
  89. 89. What does Maxeler do? Maxeler’s Dataflow Technology combines computation, data and connectivity to transform data intensive tasks from long overnight computations in a data center to real-time delivery of results at the source of data. The Maxeler Dataflow computing appliance model enables the next generation of algorithms and applications. 89
  90. 90. Maximum Performance Computing ProcessStart Original Application Identify code for acceleration and analyze bottlenecks Write MaxCompiler code Simulate Functions correctly? Build for Hardware Integrate with Host code Meets performance goals? Accelerated Application NO YESYES NO Transform app, architect and model performance Accelerate remaining code on CPU 90
  91. 91. Traditional (CPU) Computing Maxeler computes 30-200x faster, with 10-50x smaller physical footprint and 10-50x power efficiency because 100% of the chip is used for computation. Multiscale Dataflow Computing Pushing physical limits of computation. Only a small proportion of chip is actually used for computation, time is wasted talking to levels of cache. Solution Scalability – how it works… Maxeler Dataflow Technology (DFT) Maxeler computes in space not in time by maximising use of chip surface area. Our specialist tools shorten development and maintenance cycles. Our Dataflow Engines (DFEs) maximise data through-put - computation happens as a side effect. CPUs compute in time DFEs compute in space 91
  92. 92. for (int i =0; i < DATA_SIZE; i++) y[i]= x[i] * x[i] + 30; PCI Express Manager Chip Memory Manager (.java) x x + 30 x Manager m = new Manager(“Calc”); Kernel k = new MyKernel(); m.setKernel(k); m.setIO( link(“x", PCIE), m.addMode(modeDefault()); m.build(); link(“y", PCIE)); #include “MaxSLiCInterface.h” #include “Calc.max” Calc(x, y, DATA_SIZE) Main Memory CPU CPU Code CPU Code (.c) Maxeler Dataflow Compiler SLiC MaxelerOS DFEvar x = io.input("x", hwInt(32)); DFEvar result = x * x + 30; io.output("y", result, hwInt(32)); MyKernel (.java) int *x, *y; y x x + 30 y x 92
  93. 93. Maxeler Speed Advantage for High Performance Computing (HPC) Modelling 25x Finite Difference 60x Data Correlation 22x Smith-Waterman 16- 32x Fluid Flow 30x Imaging 29x 93
  94. 94. 1200m 1200m 1200m 1200m 1200m Generates >1GB every 10 The Oil Exploration Problem Image Courtesy of 94
  95. 95. Risk Solution Architecture Consistent, real-time, valuation and risk management calculations across all major asset classes Maxeler’s dataflow accelerated finance library provides ultra high speed computation of PV and risk Client provides trade, market and static data in own format Finance appliance covers 10 asset classes Risk summarizations in hardware avoid use of complex databases95
  96. 96.  Trade capture  Deal entry, storage and retrieval  Trade lifecycle management  Deal modification – new, amends and deletes  Pricing and risk management  Deal valuation  Portfolio valuation and risk  Enterprise level regulatory risk Trade Capture Java GUI Trade Lifecycle Risk engine Maxeler finance library & OS Dataflow Engines Risk and P&L results database Results of scenarios stored into client database and retrieved using client tools Results transmitted to down- stream reporting systems Results viewable in any format – e.g. Excel, Python, Matlab etc Flexible Python scripting enables rapid scenario development and deployment Trading system infrastructure - overview 96
  97. 97. Maxeler Dataflow Engines (DFEs) High Density DFEs Intel Xeon CPU cores and up to 6 DFEs with 288GB of RAM The Dataflow Appliance Dense compute with 8 DFEs, 384GB of RAM and dynamic allocation of DFEs to CPU servers with zero-copy RDMA access The Low Latency Appliance Intel Xeon CPUs and 1-2 DFEs with direct links to up to six 10Gbit Ethernet connections MaxWorkstation Desktop dataflow development system Dataflow Engines 48GB DDR3, high-speed connectivity and dense configurable logic MaxRack 10, 20 or 40 node rack systems integrating compute, networking & storage MaxCloud Hosted, on-demand, scalable accelerated compute 97
  98. 98. Maxeler University Program Members 98

    Be the first to comment

    Login to see the comments

HPC and Big Data 2013

Views

Total views

238

On Slideshare

0

From embeds

0

Number of embeds

2

Actions

Downloads

5

Shares

0

Comments

0

Likes

0

×