Tzu-Hsien Wu, Hao Shyng, J. Chou, Bin Dong, Kesheng Wu
Scientific discoveries are increasingly relying on analysis of massive amounts of data generated from scientific experiments, observations, and simulations. The ability to directly access the most relevant data records, without shifting through all of them becomes essential. While many indexing techniques have been developed to quickly locate the selected data records, the time and space required for building and storing these indexes are often too expensive to meet the demands of in situ or real-time data analysis. Existing indexing methods generally capture information about each individual data record, however, when reading a data record, the I/O system typically has to access a block or a page of data. In this work, we postulate that indexing blocks instead of individual data records could significantly reduce index size and index building time without increasing the I/O time for accessing the selected data records. Our experiments using multiple real datasets on a supercomputer show that block index can reduce query time by a factor of 2 to 50 over other existing methods, including SciDB and FastQuery. But the size of block index is almost negligible comparing to the data size, and the time of building index can reach the peak I/O speed.
{"title":"Indexing Blocks to Reduce Space and Time Requirements for Searching Large Data Files","authors":"Tzu-Hsien Wu, Hao Shyng, J. Chou, Bin Dong, Kesheng Wu","doi":"10.1109/CCGrid.2016.18","DOIUrl":"https://doi.org/10.1109/CCGrid.2016.18","url":null,"abstract":"Scientific discoveries are increasingly relying on analysis of massive amounts of data generated from scientific experiments, observations, and simulations. The ability to directly access the most relevant data records, without shifting through all of them becomes essential. While many indexing techniques have been developed to quickly locate the selected data records, the time and space required for building and storing these indexes are often too expensive to meet the demands of in situ or real-time data analysis. Existing indexing methods generally capture information about each individual data record, however, when reading a data record, the I/O system typically has to access a block or a page of data. In this work, we postulate that indexing blocks instead of individual data records could significantly reduce index size and index building time without increasing the I/O time for accessing the selected data records. Our experiments using multiple real datasets on a supercomputer show that block index can reduce query time by a factor of 2 to 50 over other existing methods, including SciDB and FastQuery. But the size of block index is almost negligible comparing to the data size, and the time of building index can reach the peak I/O speed.","PeriodicalId":103641,"journal":{"name":"2016 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid)","volume":"94 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126227386","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}
Bogdan Nicolae, Carlos H. A. Costa, Claudia Misale, K. Katrinis, Yoonho Park
Big data analytics is an indispensable tool in transforming science, engineering, medicine, healthcare, finance and ultimately business itself. With the explosion of data sizes and need for shorter time-to-solution, in-memory platforms such as Apache Spark gain increasing popularity. However, this introduces important challenges, among which data shuffling is particularly difficult: on one hand it is a key part of the computation that has a major impact on the overall performance and scalability so its efficiency is paramount, while on the other hand it needs to operate with scarce memory in order to leave as much memory available for data caching. In this context, efficient scheduling of data transfers such that it addresses both dimensions of the problem simultaneously is non-trivial. State-of-the-art solutions often rely on simple approaches that yield sub optimal performance and resource usage. This paper contributes a novel shuffle data transfer strategy that dynamically adapts to the computation with minimal memory utilization, which we briefly underline as a series of design principles.
{"title":"Towards Memory-Optimized Data Shuffling Patterns for Big Data Analytics","authors":"Bogdan Nicolae, Carlos H. A. Costa, Claudia Misale, K. Katrinis, Yoonho Park","doi":"10.1109/CCGrid.2016.85","DOIUrl":"https://doi.org/10.1109/CCGrid.2016.85","url":null,"abstract":"Big data analytics is an indispensable tool in transforming science, engineering, medicine, healthcare, finance and ultimately business itself. With the explosion of data sizes and need for shorter time-to-solution, in-memory platforms such as Apache Spark gain increasing popularity. However, this introduces important challenges, among which data shuffling is particularly difficult: on one hand it is a key part of the computation that has a major impact on the overall performance and scalability so its efficiency is paramount, while on the other hand it needs to operate with scarce memory in order to leave as much memory available for data caching. In this context, efficient scheduling of data transfers such that it addresses both dimensions of the problem simultaneously is non-trivial. State-of-the-art solutions often rely on simple approaches that yield sub optimal performance and resource usage. This paper contributes a novel shuffle data transfer strategy that dynamically adapts to the computation with minimal memory utilization, which we briefly underline as a series of design principles.","PeriodicalId":103641,"journal":{"name":"2016 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid)","volume":"66 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114893915","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}
Omer Subasi, S. Di, L. Bautista-Gomez, Prasanna Balaprakash, O. Unsal, Jesús Labarta, A. Cristal, F. Cappello
As the exascale era approaches, the increasing capacity of high-performance computing (HPC) systems with targeted power and energy budget goals introduces significant challenges in reliability. Silent data corruptions (SDCs) or silent errors are one of the major sources that corrupt the executionresults of HPC applications without being detected. In this work, we explore a low-memory-overhead SDC detector, by leveraging epsilon-insensitive support vector machine regression, to detect SDCs that occur in HPC applications that can be characterized by an impact error bound. The key contributions are three fold. (1) Our design takes spatialfeatures (i.e., neighbouring data values for each data point in a snapshot) into training data, such that little memory overhead (less than 1%) is introduced. (2) We provide an in-depth study on the detection ability and performance with different parameters, and we optimize the detection range carefully. (3) Experiments with eight real-world HPC applications show thatour detector can achieve the detection sensitivity (i.e., recall) up to 99% yet suffer a less than 1% of false positive rate for most cases. Our detector incurs low performance overhead, 5% on average, for all benchmarks studied in the paper. Compared with other state-of-the-art techniques, our detector exhibits the best tradeoff considering the detection ability and overheads.
{"title":"Spatial Support Vector Regression to Detect Silent Errors in the Exascale Era","authors":"Omer Subasi, S. Di, L. Bautista-Gomez, Prasanna Balaprakash, O. Unsal, Jesús Labarta, A. Cristal, F. Cappello","doi":"10.1109/CCGrid.2016.33","DOIUrl":"https://doi.org/10.1109/CCGrid.2016.33","url":null,"abstract":"As the exascale era approaches, the increasing capacity of high-performance computing (HPC) systems with targeted power and energy budget goals introduces significant challenges in reliability. Silent data corruptions (SDCs) or silent errors are one of the major sources that corrupt the executionresults of HPC applications without being detected. In this work, we explore a low-memory-overhead SDC detector, by leveraging epsilon-insensitive support vector machine regression, to detect SDCs that occur in HPC applications that can be characterized by an impact error bound. The key contributions are three fold. (1) Our design takes spatialfeatures (i.e., neighbouring data values for each data point in a snapshot) into training data, such that little memory overhead (less than 1%) is introduced. (2) We provide an in-depth study on the detection ability and performance with different parameters, and we optimize the detection range carefully. (3) Experiments with eight real-world HPC applications show thatour detector can achieve the detection sensitivity (i.e., recall) up to 99% yet suffer a less than 1% of false positive rate for most cases. Our detector incurs low performance overhead, 5% on average, for all benchmarks studied in the paper. Compared with other state-of-the-art techniques, our detector exhibits the best tradeoff considering the detection ability and overheads.","PeriodicalId":103641,"journal":{"name":"2016 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid)","volume":"149 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114644263","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}
Data grids are used in large scale scientific experiments to access and store nontrivial amounts of data by combining the storage resources from multiple data centers in one system. This enables users and automated services to use the storage resources in a common and efficient way. However, as data grids grow it becomes a hard problem for developers and operators to estimate how modifications in policy, hardware, and software affect the performance metrics of the data grid. In this paper we address the modeling of operational data grids. We first analyze the data grid middleware system of the ATLAS experiment at the Large Hadron Collider to identify components relevant to the data grid performance. We describe existing modeling approaches for pre-transfer, network, storage, and validation components, and build black-box models for these components. Consequently, we present a novel hybrid model, which unifies these separate component models, and we evaluate the model using an event simulator. The evaluation is based on historic workloads extracted from the ATLAS data grid. The median evaluation error of the hybrid model is at 22%.
{"title":"A Hybrid Simulation Model for Data Grids","authors":"M. Barisits, E. Kühn, M. Lassnig","doi":"10.1109/CCGrid.2016.36","DOIUrl":"https://doi.org/10.1109/CCGrid.2016.36","url":null,"abstract":"Data grids are used in large scale scientific experiments to access and store nontrivial amounts of data by combining the storage resources from multiple data centers in one system. This enables users and automated services to use the storage resources in a common and efficient way. However, as data grids grow it becomes a hard problem for developers and operators to estimate how modifications in policy, hardware, and software affect the performance metrics of the data grid. In this paper we address the modeling of operational data grids. We first analyze the data grid middleware system of the ATLAS experiment at the Large Hadron Collider to identify components relevant to the data grid performance. We describe existing modeling approaches for pre-transfer, network, storage, and validation components, and build black-box models for these components. Consequently, we present a novel hybrid model, which unifies these separate component models, and we evaluate the model using an event simulator. The evaluation is based on historic workloads extracted from the ATLAS data grid. The median evaluation error of the hybrid model is at 22%.","PeriodicalId":103641,"journal":{"name":"2016 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid)","volume":"141 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123017274","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 current era of Big Data, high volumes of a wide variety of valuable data can be easily collected and generated from a broad range of data sources of different veracities at a high velocity. Due to the well-known 5V's of these Big Data, many traditional data management approaches may not be suitable for handling the Big Data. Over the past few years, several applications and systems have developed to use cluster, cloud or grid computing to manage Big Data so as to support data science, Big Data analytics, as well as knowledge discovery and data mining. In this paper, we focus on distributed Big Data management. Specifically, we present our method for Big Data representation and management of distributed Big Data from social networks. We represent such big graph data in distributed settings so as to support big data mining of frequently occurring patterns from social networks.
{"title":"Management of Distributed Big Data for Social Networks","authors":"C. Leung, Hao Zhang","doi":"10.1109/CCGrid.2016.107","DOIUrl":"https://doi.org/10.1109/CCGrid.2016.107","url":null,"abstract":"In the current era of Big Data, high volumes of a wide variety of valuable data can be easily collected and generated from a broad range of data sources of different veracities at a high velocity. Due to the well-known 5V's of these Big Data, many traditional data management approaches may not be suitable for handling the Big Data. Over the past few years, several applications and systems have developed to use cluster, cloud or grid computing to manage Big Data so as to support data science, Big Data analytics, as well as knowledge discovery and data mining. In this paper, we focus on distributed Big Data management. Specifically, we present our method for Big Data representation and management of distributed Big Data from social networks. We represent such big graph data in distributed settings so as to support big data mining of frequently occurring patterns from social networks.","PeriodicalId":103641,"journal":{"name":"2016 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid)","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123474310","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}
Kernel-based neural networks are popular machine learning approach with many successful applications. Regularization networks represent a their special subclass with solid theoretical background and a variety of learning possibilities. In this paper, we focus on single and multi-kernel units, in particular, we describe the architecture of a product unit network, and describe an evolutionary learning algorithm for setting its parameters including different kernels from a dictionary, and optimal split of inputs into individual products. The approach is tested on real-world data from calibration of air-pollution sensor networks, and the performance is compared to several different regression tools.
{"title":"Sensor Data Air Pollution Prediction by Kernel Models","authors":"P. Vidnerová, Roman Neruda","doi":"10.1109/CCGrid.2016.80","DOIUrl":"https://doi.org/10.1109/CCGrid.2016.80","url":null,"abstract":"Kernel-based neural networks are popular machine learning approach with many successful applications. Regularization networks represent a their special subclass with solid theoretical background and a variety of learning possibilities. In this paper, we focus on single and multi-kernel units, in particular, we describe the architecture of a product unit network, and describe an evolutionary learning algorithm for setting its parameters including different kernels from a dictionary, and optimal split of inputs into individual products. The approach is tested on real-world data from calibration of air-pollution sensor networks, and the performance is compared to several different regression tools.","PeriodicalId":103641,"journal":{"name":"2016 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid)","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128081459","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}
Matheus Santos, Wagner Meira Jr, D. Guedes, Virgílio A. F. Almeida
With the recent accelerated increase in the amount of social data available in the Internet, several big data distributed processing frameworks have been proposed and implemented. Hadoop has been used widely to process all kinds of data, not only from social media. Spark is gaining popularity for offering a more flexible, object-functional, programming interface, and also by improving performance in many cases. However, not all data analysis algorithms perform well on Hadoop or Spark. For instance, graph algorithms tend to generate large amounts of messages between processing elements, which may result in poor performance even in Spark. We introduce Faster, a low latency distributed processing framework, designed to explore data locality to reduce processing costs in such algorithms. It offers an API similar to Spark, but with a slightly different execution model and new operators. Our results show that it can significantly outperform Spark on large graphs, being up to one orders of magnitude faster when running PageRank in a partial Google+ friendship graph with more than one billion edges.
{"title":"Faster: A Low Overhead Framework for Massive Data Analysis","authors":"Matheus Santos, Wagner Meira Jr, D. Guedes, Virgílio A. F. Almeida","doi":"10.1109/CCGrid.2016.90","DOIUrl":"https://doi.org/10.1109/CCGrid.2016.90","url":null,"abstract":"With the recent accelerated increase in the amount of social data available in the Internet, several big data distributed processing frameworks have been proposed and implemented. Hadoop has been used widely to process all kinds of data, not only from social media. Spark is gaining popularity for offering a more flexible, object-functional, programming interface, and also by improving performance in many cases. However, not all data analysis algorithms perform well on Hadoop or Spark. For instance, graph algorithms tend to generate large amounts of messages between processing elements, which may result in poor performance even in Spark. We introduce Faster, a low latency distributed processing framework, designed to explore data locality to reduce processing costs in such algorithms. It offers an API similar to Spark, but with a slightly different execution model and new operators. Our results show that it can significantly outperform Spark on large graphs, being up to one orders of magnitude faster when running PageRank in a partial Google+ friendship graph with more than one billion edges.","PeriodicalId":103641,"journal":{"name":"2016 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid)","volume":"76 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121114188","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}
Accelerating sequential algorithms in order to achieve high performance is often a nontrivial task. However, there are certain properties that can exacerbate this process and make it particularly daunting. For example, building an efficient parallel solution for a data-intensive algorithm requires a deep analysis of the memory access patterns and data reuse potential. Attempting to scale out the computations on clusters of machines introduces further complications due to network speed limitations. In this context, the optimization landscape can be extremely complex owing to the large number of trade-off decisions. In this paper, we discuss our experience designing two parallel implementations of an existing data-intensive machine learning algorithm that detects overlapping communities in graphs. The first design uses a single GPU to accelerate the computations of small data sets. We employed a code generation strategy in order to test and identify the best performing combination of optimizations. The second design uses a cluster of machines to scale out the computations for larger problem sizes. We used a mixture of MPI, RDMA and pipelining in order to circumvent networking overhead. Both these efforts bring us closer to understanding the complex relationships hidden within networks of entities.
{"title":"Towards Fast Overlapping Community Detection","authors":"I. El-Helw, Rutger F. H. Hofman, H. Bal","doi":"10.1109/CCGrid.2016.98","DOIUrl":"https://doi.org/10.1109/CCGrid.2016.98","url":null,"abstract":"Accelerating sequential algorithms in order to achieve high performance is often a nontrivial task. However, there are certain properties that can exacerbate this process and make it particularly daunting. For example, building an efficient parallel solution for a data-intensive algorithm requires a deep analysis of the memory access patterns and data reuse potential. Attempting to scale out the computations on clusters of machines introduces further complications due to network speed limitations. In this context, the optimization landscape can be extremely complex owing to the large number of trade-off decisions. In this paper, we discuss our experience designing two parallel implementations of an existing data-intensive machine learning algorithm that detects overlapping communities in graphs. The first design uses a single GPU to accelerate the computations of small data sets. We employed a code generation strategy in order to test and identify the best performing combination of optimizations. The second design uses a cluster of machines to scale out the computations for larger problem sizes. We used a mixture of MPI, RDMA and pipelining in order to circumvent networking overhead. Both these efforts bring us closer to understanding the complex relationships hidden within networks of entities.","PeriodicalId":103641,"journal":{"name":"2016 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid)","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125927097","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}
Accurate power estimation at runtime is essential for the efficient functioning of a power management system. While years of research have yielded accurate power models for the online prediction of instantaneous power for CPUs, such power models for graphics processing units (GPUs) are lacking. GPUs rely on low-resolution power meters that only nominally support basic power management. To address this, we propose an instantaneous power model, and in turn, a power estimator, that uses performance counters in a novel way so as to deliver accurate power estimation at runtime. Our power estimator runs on two real NVIDIA GPUs to show that accurate runtime estimation is possible without the need for the high-fidelity details that are assumed on simulation-based power models. To construct our power model, we first use correlation analysis to identify a concise set of performance counters that work well despite GPU device limitations. Next, we explore several statistical regression techniques and identify the best one. Then, to improve the prediction accuracy, we propose a novel application-dependent modeling technique, where the model is constructed online at runtime, based on the readings from a low-resolution, built-in GPU power meter. Our quantitative results show that a multi-linear model, which produces a mean absolute error of 6%, works the best in practice. An application-specific quadratic model reduces the error to nearly 1%. We show that this model can be constructed with low overhead and high accuracy at runtime. To the best of our knowledge, this is the first work attempting to model the instantaneous power of a real GPU system, earlier related work focused on average power.
{"title":"Online Power Estimation of Graphics Processing Units","authors":"Vignesh Adhinarayanan, Balaji Subramaniam, Wu-chun Feng","doi":"10.1109/CCGrid.2016.93","DOIUrl":"https://doi.org/10.1109/CCGrid.2016.93","url":null,"abstract":"Accurate power estimation at runtime is essential for the efficient functioning of a power management system. While years of research have yielded accurate power models for the online prediction of instantaneous power for CPUs, such power models for graphics processing units (GPUs) are lacking. GPUs rely on low-resolution power meters that only nominally support basic power management. To address this, we propose an instantaneous power model, and in turn, a power estimator, that uses performance counters in a novel way so as to deliver accurate power estimation at runtime. Our power estimator runs on two real NVIDIA GPUs to show that accurate runtime estimation is possible without the need for the high-fidelity details that are assumed on simulation-based power models. To construct our power model, we first use correlation analysis to identify a concise set of performance counters that work well despite GPU device limitations. Next, we explore several statistical regression techniques and identify the best one. Then, to improve the prediction accuracy, we propose a novel application-dependent modeling technique, where the model is constructed online at runtime, based on the readings from a low-resolution, built-in GPU power meter. Our quantitative results show that a multi-linear model, which produces a mean absolute error of 6%, works the best in practice. An application-specific quadratic model reduces the error to nearly 1%. We show that this model can be constructed with low overhead and high accuracy at runtime. To the best of our knowledge, this is the first work attempting to model the instantaneous power of a real GPU system, earlier related work focused on average power.","PeriodicalId":103641,"journal":{"name":"2016 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid)","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134110803","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}
Xiaodong Yu, Hao Wang, Wu-chun Feng, H. Gong, Guohua Cao
Algebraic reconstruction technique (ART) is an iterative algorithm for computed tomography (CT) image reconstruction. Due to the high computational cost, researchers turn to modern HPC systems with GPUs to accelerate the ART algorithm. However, the existing proposals suffer from inefficient designs of compressed data structure and computational kernel on GPUs. In this paper, we identify the computational patterns in the ART as the product of a sparse matrix (and its transpose) with multiple vectors (SpMV and SpMV_T). Because the implementations with well-tuned libraries, including cuSPARSE, BRC, and CSR5, underperform the expectations, we propose cuART, a complete compression and parallelization solution for the ART-based CT on GPUs. Based on the physical characteristics, i.e., the symmetries in the system matrix, we propose the symmetry-based CSR format (SCSR), which can further compress data storage by removing symmetric but redundant non-zero elements. Leveraging the sparsity patterns of X-ray projection, wetransform the CSR format to multiple dense sub-matrices in SCSR. We then design a transposition-free kernel to optimize the data access for both SpMV and SpMV_T. The experimental results illustrate that our mechanism can reduce memory usage significantly and make practical datasets fit into a single GPU. Our results also illustrate the superior performance of cuART compared to the existing methods on CPU and GPU.
{"title":"cuART: Fine-Grained Algebraic Reconstruction Technique for Computed Tomography Images on GPUs","authors":"Xiaodong Yu, Hao Wang, Wu-chun Feng, H. Gong, Guohua Cao","doi":"10.1109/CCGrid.2016.96","DOIUrl":"https://doi.org/10.1109/CCGrid.2016.96","url":null,"abstract":"Algebraic reconstruction technique (ART) is an iterative algorithm for computed tomography (CT) image reconstruction. Due to the high computational cost, researchers turn to modern HPC systems with GPUs to accelerate the ART algorithm. However, the existing proposals suffer from inefficient designs of compressed data structure and computational kernel on GPUs. In this paper, we identify the computational patterns in the ART as the product of a sparse matrix (and its transpose) with multiple vectors (SpMV and SpMV_T). Because the implementations with well-tuned libraries, including cuSPARSE, BRC, and CSR5, underperform the expectations, we propose cuART, a complete compression and parallelization solution for the ART-based CT on GPUs. Based on the physical characteristics, i.e., the symmetries in the system matrix, we propose the symmetry-based CSR format (SCSR), which can further compress data storage by removing symmetric but redundant non-zero elements. Leveraging the sparsity patterns of X-ray projection, wetransform the CSR format to multiple dense sub-matrices in SCSR. We then design a transposition-free kernel to optimize the data access for both SpMV and SpMV_T. The experimental results illustrate that our mechanism can reduce memory usage significantly and make practical datasets fit into a single GPU. Our results also illustrate the superior performance of cuART compared to the existing methods on CPU and GPU.","PeriodicalId":103641,"journal":{"name":"2016 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid)","volume":"52 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116109076","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：