Star-P is a unique technology offered by Interactive Supercomputing after nurturing at MIT. Star-P through its clever abstractions is solving the ease of use problem that has plagued supercomputing. Given that there have been around 30 parallel MATLABs including three other major offerings, Star-P must be way ahead to compete in the marketplace.Some of the innovative features of Star-P are the ability to program in MATLAB, hook in task parallel codes written using a processor free abstraction, hook in existing parallel codes, and obtain the performance that represents the HPC promise. All this is through a client/server interface. Other clients such as Python or R could be possible. The MATLAB, Python, or R becomes the "browser." If we make it look easy, it is because decades of parallel computing experience has taught us that it is not.This talk demonstrates the abstractions and innovations that make this possible.
{"title":"So What's innovative and exotic about star-P for MATLAB and other clients?","authors":"A. Edelman","doi":"10.1145/1188455.1188746","DOIUrl":"https://doi.org/10.1145/1188455.1188746","url":null,"abstract":"Star-P is a unique technology offered by Interactive Supercomputing after nurturing at MIT. Star-P through its clever abstractions is solving the ease of use problem that has plagued supercomputing. Given that there have been around 30 parallel MATLABs including three other major offerings, Star-P must be way ahead to compete in the marketplace.Some of the innovative features of Star-P are the ability to program in MATLAB, hook in task parallel codes written using a processor free abstraction, hook in existing parallel codes, and obtain the performance that represents the HPC promise. All this is through a client/server interface. Other clients such as Python or R could be possible. The MATLAB, Python, or R becomes the \"browser.\" If we make it look easy, it is because decades of parallel computing experience has taught us that it is not.This talk demonstrates the abstractions and innovations that make this possible.","PeriodicalId":115940,"journal":{"name":"Proceedings of the 2006 ACM/IEEE conference on Supercomputing","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123039183","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
D. Luebke, Mark J. Harris, N. Govindaraju, A. Lefohn, M. Houston, John Douglas Owens, Mark E. Segal, Matthew Papakipos, I. Buck
The graphics processor (GPU) on today's commodity video cards has evolved into an extremely powerful and flexible processor. Modern graphics architectures provide tremendous memory bandwidth and computational horsepower, with dozens of fully programmable shading units that support vector operations and IEEE floating point precision. High-level languages have emerged for graphics hardware, making this computational power accessible. GPGPU stands for "General-Purpose Computation on GPUs". GPGPU researchers have achieved over an order of magnitude speedup over modern CPUs on some non-graphics problems.This course provides detailed coverage of general-purpose computation on graphics hardware. We emphasize core computational building blocks, ranging from linear algebra to database queries, and review the tools, perils, and strategies in GPU programming. We present analysis of GPU performance characteristics, and use this analysis to provide insight into how to build efficient GPGPU algorithms. Finally we present a set of case studies on general-purpose applications of graphics hardware.
{"title":"GPGPU: general-purpose computation on graphics hardware","authors":"D. Luebke, Mark J. Harris, N. Govindaraju, A. Lefohn, M. Houston, John Douglas Owens, Mark E. Segal, Matthew Papakipos, I. Buck","doi":"10.1145/1188455.1188672","DOIUrl":"https://doi.org/10.1145/1188455.1188672","url":null,"abstract":"The graphics processor (GPU) on today's commodity video cards has evolved into an extremely powerful and flexible processor. Modern graphics architectures provide tremendous memory bandwidth and computational horsepower, with dozens of fully programmable shading units that support vector operations and IEEE floating point precision. High-level languages have emerged for graphics hardware, making this computational power accessible. GPGPU stands for \"General-Purpose Computation on GPUs\". GPGPU researchers have achieved over an order of magnitude speedup over modern CPUs on some non-graphics problems.This course provides detailed coverage of general-purpose computation on graphics hardware. We emphasize core computational building blocks, ranging from linear algebra to database queries, and review the tools, perils, and strategies in GPU programming. We present analysis of GPU performance characteristics, and use this analysis to provide insight into how to build efficient GPGPU algorithms. Finally we present a set of case studies on general-purpose applications of graphics hardware.","PeriodicalId":115940,"journal":{"name":"Proceedings of the 2006 ACM/IEEE conference on Supercomputing","volume":"309 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123142733","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The paradigm shift rate is now doubling every decade, so the twenty-first century will see 20,000 years of progress at today's rate. Computation, communication, biological technologies (for example, DNA sequencing), brain scanning, knowledge of the human brain, and human knowledge in general are all accelerating at an even faster pace, generally doubling price-performance, capacity, and bandwidth every year. The well-known Moore's Law is only one example of many of this inherent acceleration. The size of the key features of technology is also shrinking, at a rate of about 4 per linear dimension per decade. Three-dimensional molecular computing will provide the hardware for human-level "strong" AI well before 2030. The more important software insights will be gained in part from the reverse-engineering of the human brain, a process well under way.We are rapidly learning the software programs called genes that underlie biology. We are understanding disease and aging processes as information processes, and are gaining the tools to reprogram them. RNA interference, for example, allows us to turn selected genes off, and new forms of gene therapy are enabling us to effectively add new genes. Within one to two decades, we will be in a position to stop and reverse the progression of disease and aging resulting in dramatic gains in health and longevity.The fraction of value of products and services comprised by software and related forms of information is rapidly asymptoting to 100 percent The deflation rate for information technologies, both hardware and software, is about 50 percent per year, providing a powerful deflationary force in the economy. The portion of the economy comprised of information technology is itself growing exponentially and within a couple of decades, the bulk of the economy will be dominated by information and software.Once nonbiological intelligence matches the range and subtlety of human intelligence, it will necessarily soar past it because of the continuing acceleration of information-based technologies, as well as the ability of machines to instantly share their knowledge. Intelligent nanorobots will be deeply integrated in the environment, our bodies and our brains, providing vastly extended longevity, full-immersion virtual reality incorporating all of the senses, experience "beaming," and enhanced human intelligence. The implication will be an intimate merger between the technology-creating species and the evolutionary process it spawned.
{"title":"The coming merger of biological and non biological intelligence","authors":"R. Kurzweil","doi":"10.1145/1188455.1188658","DOIUrl":"https://doi.org/10.1145/1188455.1188658","url":null,"abstract":"The paradigm shift rate is now doubling every decade, so the twenty-first century will see 20,000 years of progress at today's rate. Computation, communication, biological technologies (for example, DNA sequencing), brain scanning, knowledge of the human brain, and human knowledge in general are all accelerating at an even faster pace, generally doubling price-performance, capacity, and bandwidth every year. The well-known Moore's Law is only one example of many of this inherent acceleration. The size of the key features of technology is also shrinking, at a rate of about 4 per linear dimension per decade. Three-dimensional molecular computing will provide the hardware for human-level \"strong\" AI well before 2030. The more important software insights will be gained in part from the reverse-engineering of the human brain, a process well under way.We are rapidly learning the software programs called genes that underlie biology. We are understanding disease and aging processes as information processes, and are gaining the tools to reprogram them. RNA interference, for example, allows us to turn selected genes off, and new forms of gene therapy are enabling us to effectively add new genes. Within one to two decades, we will be in a position to stop and reverse the progression of disease and aging resulting in dramatic gains in health and longevity.The fraction of value of products and services comprised by software and related forms of information is rapidly asymptoting to 100 percent The deflation rate for information technologies, both hardware and software, is about 50 percent per year, providing a powerful deflationary force in the economy. The portion of the economy comprised of information technology is itself growing exponentially and within a couple of decades, the bulk of the economy will be dominated by information and software.Once nonbiological intelligence matches the range and subtlety of human intelligence, it will necessarily soar past it because of the continuing acceleration of information-based technologies, as well as the ability of machines to instantly share their knowledge. Intelligent nanorobots will be deeply integrated in the environment, our bodies and our brains, providing vastly extended longevity, full-immersion virtual reality incorporating all of the senses, experience \"beaming,\" and enhanced human intelligence. The implication will be an intimate merger between the technology-creating species and the evolutionary process it spawned.","PeriodicalId":115940,"journal":{"name":"Proceedings of the 2006 ACM/IEEE conference on Supercomputing","volume":"17 3","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121001000","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
SPEC HPG has been creating benchmarks for comparing HPC systems. Currently they have a benchmark suite based on OpenMP (SPEC OMP) and another based on real applications supporting multiple parallel paradigms (SPEC HPC2002). SPEC is now in the process of developing a benhcmark based on MPI applications. This new suite is expected to measure MPI, CPU, compiler, memory bandwidth and interconnect performance. The BOF will focus on SPEC HPG benchmarks and will introduce the new SPEC MPI2006 benchmark.
{"title":"SPEC HPG benchmarks","authors":"K. Kalyanasundaram","doi":"10.1145/1188455.1188473","DOIUrl":"https://doi.org/10.1145/1188455.1188473","url":null,"abstract":"SPEC HPG has been creating benchmarks for comparing HPC systems. Currently they have a benchmark suite based on OpenMP (SPEC OMP) and another based on real applications supporting multiple parallel paradigms (SPEC HPC2002). SPEC is now in the process of developing a benhcmark based on MPI applications. This new suite is expected to measure MPI, CPU, compiler, memory bandwidth and interconnect performance. The BOF will focus on SPEC HPG benchmarks and will introduce the new SPEC MPI2006 benchmark.","PeriodicalId":115940,"journal":{"name":"Proceedings of the 2006 ACM/IEEE conference on Supercomputing","volume":"121 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126048410","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The Canadian Design Research Network-(CDRN) is setting up a prototype Design Grid. The Design Grid has several goals including:1)enabling national and international design reviews in areas ranging from urban planning, landscape architecture, architect and aerospace engineering. 2)allowing interactive discussions including video/audio streaming and shared powerpoints between highly distributed individuals and institutions where individuals may only have access to very modest technical resources. 3)allowing the public to access visualizations of buildings and see the impact of those buildings. 4)allow the sharing of design courses across universities.
{"title":"The Canadian design research network's prototype design grid","authors":"J. Danahy, West Suhanic","doi":"10.1145/1188455.1188788","DOIUrl":"https://doi.org/10.1145/1188455.1188788","url":null,"abstract":"The Canadian Design Research Network-(CDRN) is setting up a prototype Design Grid. The Design Grid has several goals including:1)enabling national and international design reviews in areas ranging from urban planning, landscape architecture, architect and aerospace engineering. 2)allowing interactive discussions including video/audio streaming and shared powerpoints between highly distributed individuals and institutions where individuals may only have access to very modest technical resources. 3)allowing the public to access visualizations of buildings and see the impact of those buildings. 4)allow the sharing of design courses across universities.","PeriodicalId":115940,"journal":{"name":"Proceedings of the 2006 ACM/IEEE conference on Supercomputing","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125507725","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In the coming decade, the visible universe will be surveyed to unprecedented breadth, depth, and spectral coverage to explore its structures, past history, and future fate. Data volumes of many petabytes are expected. Numerical simulations of cosmic structure formation and evolution must keep pace in scale and complexity if they are to assist in the interpretation of such data, as well as to make predictions. We describe the Cosmic Simulator, a software facility under development at UCSD for performing, analyzing, and archiving cosmological simulations of unprecedented scale and physical realism. Its major components are the adaptive mesh refinement hydrodynamic cosmology simulation code ENZO, a data analysis pipeline, and a Storage Resource Broker (SRB)-managed data archive housed at SDSC. We show scientific results from the Cosmic Simulator implemented on a data grid connecting LLNL and SDSC computational resources, and discuss future plans.
{"title":"The cosmic simulator","authors":"M. Norman","doi":"10.1145/1188455.1188509","DOIUrl":"https://doi.org/10.1145/1188455.1188509","url":null,"abstract":"In the coming decade, the visible universe will be surveyed to unprecedented breadth, depth, and spectral coverage to explore its structures, past history, and future fate. Data volumes of many petabytes are expected. Numerical simulations of cosmic structure formation and evolution must keep pace in scale and complexity if they are to assist in the interpretation of such data, as well as to make predictions. We describe the Cosmic Simulator, a software facility under development at UCSD for performing, analyzing, and archiving cosmological simulations of unprecedented scale and physical realism. Its major components are the adaptive mesh refinement hydrodynamic cosmology simulation code ENZO, a data analysis pipeline, and a Storage Resource Broker (SRB)-managed data archive housed at SDSC. We show scientific results from the Cosmic Simulator implemented on a data grid connecting LLNL and SDSC computational resources, and discuss future plans.","PeriodicalId":115940,"journal":{"name":"Proceedings of the 2006 ACM/IEEE conference on Supercomputing","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114296402","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Parallel Computing Toolkit is a Mathematica package that allows users to develop and run parallel computations. It provides a high-level language for expressing parallel algorithms in a machine-independent way supporting many standard parallel programming paradigms. gridMathematica takes full advantage of Parallel Computing Toolkit for providing an easy-to-use environment for parallel computations on multi-processor machines, clusters, or grids of compute nodes using the scale and power of Mathematica. Among the new features under development is an extension of the Wolfram Workbench (which is based on Eclipse) that allows development, debugging and profiling of parallel Mathematica applications. The presentation discusses the features of gridMathematica and demonstrates examples of real world applications currently being used today. Attendees are encouraged to bring their problems to the workshop for evaluation and suggestion.
{"title":"gridMathematica: overview and new developments","authors":"Roman Maeder","doi":"10.1145/1188455.1188720","DOIUrl":"https://doi.org/10.1145/1188455.1188720","url":null,"abstract":"Parallel Computing Toolkit is a Mathematica package that allows users to develop and run parallel computations. It provides a high-level language for expressing parallel algorithms in a machine-independent way supporting many standard parallel programming paradigms. gridMathematica takes full advantage of Parallel Computing Toolkit for providing an easy-to-use environment for parallel computations on multi-processor machines, clusters, or grids of compute nodes using the scale and power of Mathematica. Among the new features under development is an extension of the Wolfram Workbench (which is based on Eclipse) that allows development, debugging and profiling of parallel Mathematica applications. The presentation discusses the features of gridMathematica and demonstrates examples of real world applications currently being used today. Attendees are encouraged to bring their problems to the workshop for evaluation and suggestion.","PeriodicalId":115940,"journal":{"name":"Proceedings of the 2006 ACM/IEEE conference on Supercomputing","volume":"47 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117299685","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
QsNetIII and 10 Gbit/s Ethernet, two networks for high perfomance computing. While the proprietary QsNet will continue to provide supercomputing funcionalities to clusters of commodity based servers, the second will establish itself as preferred choice in capacity class systems.Quadrics, who is actively developing products based on the two technologies, will present early performance results and comparisons.
{"title":"High performance networks for the future","authors":"D. Roweth, M. McLaren","doi":"10.1145/1188455.1188728","DOIUrl":"https://doi.org/10.1145/1188455.1188728","url":null,"abstract":"QsNetIII and 10 Gbit/s Ethernet, two networks for high perfomance computing. While the proprietary QsNet will continue to provide supercomputing funcionalities to clusters of commodity based servers, the second will establish itself as preferred choice in capacity class systems.Quadrics, who is actively developing products based on the two technologies, will present early performance results and comparisons.","PeriodicalId":115940,"journal":{"name":"Proceedings of the 2006 ACM/IEEE conference on Supercomputing","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129411889","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We are on the verge of a paradigm shift with our software for the new multicore architectures and there is no free lunch for conventional software. Power consumption and heat dissipation issues are pushing the microprocessor industry towards multicore design patterns. With the number of cores on multicore chips expected to reach tens to perhaps hundreds in a few years, efficient implementations of numerical libraries using shared memory programming models is of high interest. The current message passing paradigm used in ScaLAPACK and elsewhere introduces unnecessary memory overhead and memory copy operations, which degrade performance, along with making it harder to schedule operations that could be done in parallel. Limiting the use of shared memory to fork-join parallelism (perhaps with OpenMP) or to focusing the parallelism within the BLAS does not address all these issues.
{"title":"Targeting multi-core architectures for linear algebra applications","authors":"A. Buttari, J. Kurzak, J. Dongarra","doi":"10.1145/1188455.1188623","DOIUrl":"https://doi.org/10.1145/1188455.1188623","url":null,"abstract":"We are on the verge of a paradigm shift with our software for the new multicore architectures and there is no free lunch for conventional software. Power consumption and heat dissipation issues are pushing the microprocessor industry towards multicore design patterns. With the number of cores on multicore chips expected to reach tens to perhaps hundreds in a few years, efficient implementations of numerical libraries using shared memory programming models is of high interest. The current message passing paradigm used in ScaLAPACK and elsewhere introduces unnecessary memory overhead and memory copy operations, which degrade performance, along with making it harder to schedule operations that could be done in parallel. Limiting the use of shared memory to fork-join parallelism (perhaps with OpenMP) or to focusing the parallelism within the BLAS does not address all these issues.","PeriodicalId":115940,"journal":{"name":"Proceedings of the 2006 ACM/IEEE conference on Supercomputing","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129140176","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We describe a new environment for large-scale turbulence simulations that uses a cluster of database nodes to store the complete space-time history of fluid velocities. This allows for rapid access to high resolution data that were traditionally too large to store and too computationally expensive to produce on demand.We perform the actual experimental analysis inside the database nodes, which allows for data-intensive computations to be performed across a large number of nodes with relatively little network traffic.We currently have a limited-scale prototype system running actual turbulence simulations and are in the process of establishing a production cluster with high-resolution data. We will discuss our design choices and initial results with load balancing a data-intensive, migratory workload.
{"title":"Engineering the 100 terabyte turbulence database (or how to track particles at home)","authors":"E. Perlman, R. Burns","doi":"10.1145/1188455.1188625","DOIUrl":"https://doi.org/10.1145/1188455.1188625","url":null,"abstract":"We describe a new environment for large-scale turbulence simulations that uses a cluster of database nodes to store the complete space-time history of fluid velocities. This allows for rapid access to high resolution data that were traditionally too large to store and too computationally expensive to produce on demand.We perform the actual experimental analysis inside the database nodes, which allows for data-intensive computations to be performed across a large number of nodes with relatively little network traffic.We currently have a limited-scale prototype system running actual turbulence simulations and are in the process of establishing a production cluster with high-resolution data. We will discuss our design choices and initial results with load balancing a data-intensive, migratory workload.","PeriodicalId":115940,"journal":{"name":"Proceedings of the 2006 ACM/IEEE conference on Supercomputing","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129916728","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：