The Experts below are selected from a list of 4857 Experts worldwide ranked by ideXlab platform
Roy H. Campbell - One of the best experts on this subject based on the ideXlab platform.
-
CLUSTER - PIC: Partitioned Iterative Convergence for Clusters
2012 IEEE International Conference on Cluster Computing, 2012Co-Authors: Reza Farivar, Anand Raghunathan, Srimat Chakradhar, Harshit Kharbanda, Roy H. CampbellAbstract:Iterative-Convergence algorithms are frequently used in a variety of domains to build models from large data sets. Cluster implementations of these algorithms are commonly realized using parallel programming models such as MapReduce. However, these implementations suffer from significant performance bottlenecks, especially due to large volumes of network traffic resulting from intermediate data and model updates during the iterations. To address these challenges, we propose partitioned Iterative Convergence (PIC), a new approach to programming and executing Iterative Convergence algorithms on frameworks like MapReduce. In PIC, we execute the Iterative-Convergence computation in two phases - the best-effort phase, which quickly produces a good initial model and the top-off phase, which further refines this model to produce the final solution. The best-effort phase Iteratively performs the following steps: (a) partition the input data and the model to create several smaller, model-building sub-problems, (b) independently solve these sub-problems using Iterative Convergence computations, and (c) merge solutions of the sub-problems to create the next version of the model. This partitioned, loosely coupled execution of the computation produces a model of good quality, while drastically reducing network traffic due to intermediate data and model updates. The top-off phase further refines this model by employing the original Iterative-Convergence computation on the entire (un-partitioned) problem until Convergence. However, the number of iterations executed in the top-off phase is quite small, resulting in a significant overall improvement in performance. We have implemented a library for PIC on top of the Hadoop MapReduce framework, and evaluated it using five popular Iterative-Convergence algorithms (Page Rank, K-Means clustering, neural network training, linear equation solver and image smoothing). Our evaluations on clusters ranging from 6 nodes to 256 nodes demonstrate a 2.5X-4X speedup compared to conventional implementations using Hadoop.
-
PIC: Partitioned Iterative Convergence for Clusters
2012 IEEE International Conference on Cluster Computing, 2012Co-Authors: Reza Farivar, Anand Raghunathan, Srimat Chakradhar, Harshit Kharbanda, Roy H. CampbellAbstract:Iterative-Convergence algorithms are frequently used in a variety of domains to build models from large data sets. Cluster implementations of these algorithms are commonly realized using parallel programming models such as MapReduce. However, these implementations suffer from significant performance bottlenecks, especially due to large volumes of network traffic resulting from intermediate data and model updates during the iterations. To address these challenges, we propose partitioned Iterative Convergence (PIC), a new approach to programming and executing Iterative Convergence algorithms on frameworks like MapReduce. In PIC, we execute the Iterative-Convergence computation in two phases - the best-effort phase, which quickly produces a good initial model and the top-off phase, which further refines this model to produce the final solution. The best-effort phase Iteratively performs the following steps: (a) partition the input data and the model to create several smaller, model-building sub-problems, (b) independently solve these sub-problems using Iterative Convergence computations, and (c) merge solutions of the sub-problems to create the next version of the model. This partitioned, loosely coupled execution of the computation produces a model of good quality, while drastically reducing network traffic due to intermediate data and model updates. The top-off phase further refines this model by employing the original Iterative-Convergence computation on the entire (un-partitioned) problem until Convergence. However, the number of iterations executed in the top-off phase is quite small, resulting in a significant overall improvement in performance. We have implemented a library for PIC on top of the Hadoop MapReduce framework, and evaluated it using five popular Iterative-Convergence algorithms (Page Rank, K-Means clustering, neural network training, linear equation solver and image smoothing). Our evaluations on clusters ranging from 6 nodes to 256 nodes demonstrate a 2.5X-4X speedup compared to conventional implementations using Hadoop.
Reza Farivar - One of the best experts on this subject based on the ideXlab platform.
-
CLUSTER - PIC: Partitioned Iterative Convergence for Clusters
2012 IEEE International Conference on Cluster Computing, 2012Co-Authors: Reza Farivar, Anand Raghunathan, Srimat Chakradhar, Harshit Kharbanda, Roy H. CampbellAbstract:Iterative-Convergence algorithms are frequently used in a variety of domains to build models from large data sets. Cluster implementations of these algorithms are commonly realized using parallel programming models such as MapReduce. However, these implementations suffer from significant performance bottlenecks, especially due to large volumes of network traffic resulting from intermediate data and model updates during the iterations. To address these challenges, we propose partitioned Iterative Convergence (PIC), a new approach to programming and executing Iterative Convergence algorithms on frameworks like MapReduce. In PIC, we execute the Iterative-Convergence computation in two phases - the best-effort phase, which quickly produces a good initial model and the top-off phase, which further refines this model to produce the final solution. The best-effort phase Iteratively performs the following steps: (a) partition the input data and the model to create several smaller, model-building sub-problems, (b) independently solve these sub-problems using Iterative Convergence computations, and (c) merge solutions of the sub-problems to create the next version of the model. This partitioned, loosely coupled execution of the computation produces a model of good quality, while drastically reducing network traffic due to intermediate data and model updates. The top-off phase further refines this model by employing the original Iterative-Convergence computation on the entire (un-partitioned) problem until Convergence. However, the number of iterations executed in the top-off phase is quite small, resulting in a significant overall improvement in performance. We have implemented a library for PIC on top of the Hadoop MapReduce framework, and evaluated it using five popular Iterative-Convergence algorithms (Page Rank, K-Means clustering, neural network training, linear equation solver and image smoothing). Our evaluations on clusters ranging from 6 nodes to 256 nodes demonstrate a 2.5X-4X speedup compared to conventional implementations using Hadoop.
-
PIC: Partitioned Iterative Convergence for Clusters
2012 IEEE International Conference on Cluster Computing, 2012Co-Authors: Reza Farivar, Anand Raghunathan, Srimat Chakradhar, Harshit Kharbanda, Roy H. CampbellAbstract:Iterative-Convergence algorithms are frequently used in a variety of domains to build models from large data sets. Cluster implementations of these algorithms are commonly realized using parallel programming models such as MapReduce. However, these implementations suffer from significant performance bottlenecks, especially due to large volumes of network traffic resulting from intermediate data and model updates during the iterations. To address these challenges, we propose partitioned Iterative Convergence (PIC), a new approach to programming and executing Iterative Convergence algorithms on frameworks like MapReduce. In PIC, we execute the Iterative-Convergence computation in two phases - the best-effort phase, which quickly produces a good initial model and the top-off phase, which further refines this model to produce the final solution. The best-effort phase Iteratively performs the following steps: (a) partition the input data and the model to create several smaller, model-building sub-problems, (b) independently solve these sub-problems using Iterative Convergence computations, and (c) merge solutions of the sub-problems to create the next version of the model. This partitioned, loosely coupled execution of the computation produces a model of good quality, while drastically reducing network traffic due to intermediate data and model updates. The top-off phase further refines this model by employing the original Iterative-Convergence computation on the entire (un-partitioned) problem until Convergence. However, the number of iterations executed in the top-off phase is quite small, resulting in a significant overall improvement in performance. We have implemented a library for PIC on top of the Hadoop MapReduce framework, and evaluated it using five popular Iterative-Convergence algorithms (Page Rank, K-Means clustering, neural network training, linear equation solver and image smoothing). Our evaluations on clusters ranging from 6 nodes to 256 nodes demonstrate a 2.5X-4X speedup compared to conventional implementations using Hadoop.
Juan Chen - One of the best experts on this subject based on the ideXlab platform.
-
Metis-CIC: A new mesh partitioning heuristic for parallel preconditioned Iterative methods in CFD
2016 International Conference on High Performance Computing & Simulation (HPCS), 2016Co-Authors: Miao Wang, Wenjing Yang, Hao Li, Juan ChenAbstract:Mesh partitioning has a significant influence on the efficiency of parallel computational fluid dynamics (CFD). This paper considers the most widely-used preconditioned conjugated gradient (PCG) methods for solving the linear systems in parallel since it is the core and most time-consuming part in CFD. Based on the analysis of how mesh partitioning impacts the solution time of parallel PCG, we propose a new cost function combining communication overhead and Iterative Convergence rate together. An adjustable parameter is involved in the cost function. For a specific application and a given parallel degree, the parameter can be fitted using profiling information of the application. A partitioning method based on the new cost function, named Metis-CIC (Communication overhead and Iterative Convergence rate Combined), is implemented in Metis (a general graph partitioning tool). Numerical results of two CFD applications show that Metis-CIC outperforms the standard Metis partitioning routine and Metis-Acut which only considers the Iterative Convergence rate during the partitioning.
-
HPCS - Metis-CIC: A new mesh partitioning heuristic for parallel preconditioned Iterative methods in CFD
2016 International Conference on High Performance Computing & Simulation (HPCS), 2016Co-Authors: Miao Wang, Wenjing Yang, Hao Li, Juan ChenAbstract:Mesh partitioning has a significant influence on the efficiency of parallel computational fluid dynamics (CFD). This paper considers the most widely-used preconditioned conjugated gradient (PCG) methods for solving the linear systems in parallel since it is the core and most time-consuming part in CFD. Based on the analysis of how mesh partitioning impacts the solution time of parallel PCG, we propose a new cost function combining communication overhead and Iterative Convergence rate together. An adjustable parameter is involved in the cost function. For a specific application and a given parallel degree, the parameter can be fitted using profiling information of the application. A partitioning method based on the new cost function, named Metis-CIC (Communication overhead and Iterative Convergence rate Combined), is implemented in Metis (a general graph partitioning tool). Numerical results of two CFD applications show that Metis-CIC outperforms the standard Metis partitioning routine and Metis-Acut which only considers the Iterative Convergence rate during the partitioning.
-
Mesh-Partitioning Metrics for Parallel Preconditioned Conjugated Gradient Solvers in CFD
2016 3rd International Conference on Information Science and Control Engineering (ICISCE), 2016Co-Authors: Miao Wang, Hao Li, Juan ChenAbstract:This paper focuses on mesh-partitioning metrics in large-scale parallel computational fluid dynamics (CFD) simulations. Mesh partitioning has a significant influence on the efficiency of parallel preconditioned conjugated gradient (PCG) solving procedure, which is the most representative and time-consuming part in parallel CFD. As the efficiency of parallel PCG depends on load balancing, communication overhead and Iterative Convergence rate comprehensively, we present a detailed review of mesh-partitioning metrics on these three aspects respectively. Three typical large-scale CFD applications are built to numerically testify the validity of all those metrics.
Miao Wang - One of the best experts on this subject based on the ideXlab platform.
-
Metis-CIC: A new mesh partitioning heuristic for parallel preconditioned Iterative methods in CFD
2016 International Conference on High Performance Computing & Simulation (HPCS), 2016Co-Authors: Miao Wang, Wenjing Yang, Hao Li, Juan ChenAbstract:Mesh partitioning has a significant influence on the efficiency of parallel computational fluid dynamics (CFD). This paper considers the most widely-used preconditioned conjugated gradient (PCG) methods for solving the linear systems in parallel since it is the core and most time-consuming part in CFD. Based on the analysis of how mesh partitioning impacts the solution time of parallel PCG, we propose a new cost function combining communication overhead and Iterative Convergence rate together. An adjustable parameter is involved in the cost function. For a specific application and a given parallel degree, the parameter can be fitted using profiling information of the application. A partitioning method based on the new cost function, named Metis-CIC (Communication overhead and Iterative Convergence rate Combined), is implemented in Metis (a general graph partitioning tool). Numerical results of two CFD applications show that Metis-CIC outperforms the standard Metis partitioning routine and Metis-Acut which only considers the Iterative Convergence rate during the partitioning.
-
HPCS - Metis-CIC: A new mesh partitioning heuristic for parallel preconditioned Iterative methods in CFD
2016 International Conference on High Performance Computing & Simulation (HPCS), 2016Co-Authors: Miao Wang, Wenjing Yang, Hao Li, Juan ChenAbstract:Mesh partitioning has a significant influence on the efficiency of parallel computational fluid dynamics (CFD). This paper considers the most widely-used preconditioned conjugated gradient (PCG) methods for solving the linear systems in parallel since it is the core and most time-consuming part in CFD. Based on the analysis of how mesh partitioning impacts the solution time of parallel PCG, we propose a new cost function combining communication overhead and Iterative Convergence rate together. An adjustable parameter is involved in the cost function. For a specific application and a given parallel degree, the parameter can be fitted using profiling information of the application. A partitioning method based on the new cost function, named Metis-CIC (Communication overhead and Iterative Convergence rate Combined), is implemented in Metis (a general graph partitioning tool). Numerical results of two CFD applications show that Metis-CIC outperforms the standard Metis partitioning routine and Metis-Acut which only considers the Iterative Convergence rate during the partitioning.
-
Mesh-Partitioning Metrics for Parallel Preconditioned Conjugated Gradient Solvers in CFD
2016 3rd International Conference on Information Science and Control Engineering (ICISCE), 2016Co-Authors: Miao Wang, Hao Li, Juan ChenAbstract:This paper focuses on mesh-partitioning metrics in large-scale parallel computational fluid dynamics (CFD) simulations. Mesh partitioning has a significant influence on the efficiency of parallel preconditioned conjugated gradient (PCG) solving procedure, which is the most representative and time-consuming part in parallel CFD. As the efficiency of parallel PCG depends on load balancing, communication overhead and Iterative Convergence rate comprehensively, we present a detailed review of mesh-partitioning metrics on these three aspects respectively. Three typical large-scale CFD applications are built to numerically testify the validity of all those metrics.
-
DMRPar: A Dynamic Mesh Repartitioning Scheme for Dam Break Simulations in OpenFOAM
2016 17th International Conference on Parallel and Distributed Computing Applications and Technologies (PDCAT), 2016Co-Authors: Miao Wang, Chao Li, Zhiling LiAbstract:For parallel dam break simulations in OpenFOAM (Open Source Field Operation and Manipulation), the core procedure is solving linear systems using Iterative methods and the Iterative Convergence rate is significant to the overall efficiency. A dynamic mesh repartitioning scheme DMRPar (Dynamic Mesh Re-Partitioning) considering the Iterative Convergence feature is implemented in OpenFOAM. Given that the numerical characteristics of linear systems change a lot along with the complex flow field, DMRPar takes linear system information from the previous timestep into account for the repartitioning at the current timestep. The implementation reuses current mesh topology in OpenFOAM and calculates distributed adjacency graph structure for the mesh. The repartitioning heuristic is based on a general multi-level parallel graph partitioning package called ParMetis. Numerical results on two typical dam break simulations show that DMRPar outperforms the traditional static partitioning method significantly in the total simulation time.
Srimat Chakradhar - One of the best experts on this subject based on the ideXlab platform.
-
CLUSTER - PIC: Partitioned Iterative Convergence for Clusters
2012 IEEE International Conference on Cluster Computing, 2012Co-Authors: Reza Farivar, Anand Raghunathan, Srimat Chakradhar, Harshit Kharbanda, Roy H. CampbellAbstract:Iterative-Convergence algorithms are frequently used in a variety of domains to build models from large data sets. Cluster implementations of these algorithms are commonly realized using parallel programming models such as MapReduce. However, these implementations suffer from significant performance bottlenecks, especially due to large volumes of network traffic resulting from intermediate data and model updates during the iterations. To address these challenges, we propose partitioned Iterative Convergence (PIC), a new approach to programming and executing Iterative Convergence algorithms on frameworks like MapReduce. In PIC, we execute the Iterative-Convergence computation in two phases - the best-effort phase, which quickly produces a good initial model and the top-off phase, which further refines this model to produce the final solution. The best-effort phase Iteratively performs the following steps: (a) partition the input data and the model to create several smaller, model-building sub-problems, (b) independently solve these sub-problems using Iterative Convergence computations, and (c) merge solutions of the sub-problems to create the next version of the model. This partitioned, loosely coupled execution of the computation produces a model of good quality, while drastically reducing network traffic due to intermediate data and model updates. The top-off phase further refines this model by employing the original Iterative-Convergence computation on the entire (un-partitioned) problem until Convergence. However, the number of iterations executed in the top-off phase is quite small, resulting in a significant overall improvement in performance. We have implemented a library for PIC on top of the Hadoop MapReduce framework, and evaluated it using five popular Iterative-Convergence algorithms (Page Rank, K-Means clustering, neural network training, linear equation solver and image smoothing). Our evaluations on clusters ranging from 6 nodes to 256 nodes demonstrate a 2.5X-4X speedup compared to conventional implementations using Hadoop.
-
PIC: Partitioned Iterative Convergence for Clusters
2012 IEEE International Conference on Cluster Computing, 2012Co-Authors: Reza Farivar, Anand Raghunathan, Srimat Chakradhar, Harshit Kharbanda, Roy H. CampbellAbstract:Iterative-Convergence algorithms are frequently used in a variety of domains to build models from large data sets. Cluster implementations of these algorithms are commonly realized using parallel programming models such as MapReduce. However, these implementations suffer from significant performance bottlenecks, especially due to large volumes of network traffic resulting from intermediate data and model updates during the iterations. To address these challenges, we propose partitioned Iterative Convergence (PIC), a new approach to programming and executing Iterative Convergence algorithms on frameworks like MapReduce. In PIC, we execute the Iterative-Convergence computation in two phases - the best-effort phase, which quickly produces a good initial model and the top-off phase, which further refines this model to produce the final solution. The best-effort phase Iteratively performs the following steps: (a) partition the input data and the model to create several smaller, model-building sub-problems, (b) independently solve these sub-problems using Iterative Convergence computations, and (c) merge solutions of the sub-problems to create the next version of the model. This partitioned, loosely coupled execution of the computation produces a model of good quality, while drastically reducing network traffic due to intermediate data and model updates. The top-off phase further refines this model by employing the original Iterative-Convergence computation on the entire (un-partitioned) problem until Convergence. However, the number of iterations executed in the top-off phase is quite small, resulting in a significant overall improvement in performance. We have implemented a library for PIC on top of the Hadoop MapReduce framework, and evaluated it using five popular Iterative-Convergence algorithms (Page Rank, K-Means clustering, neural network training, linear equation solver and image smoothing). Our evaluations on clusters ranging from 6 nodes to 256 nodes demonstrate a 2.5X-4X speedup compared to conventional implementations using Hadoop.
