http://sedici.unlp.edu.ar:80/handle/10915/126041 2026-11-02T21:46:03Z 2026-11-02T21:46:03Z González Márquez, D. Fernández Slezak, Diego Turjanski, P. Mocskos, E. http://sedici.unlp.edu.ar:80/handle/10915/126144 2021-10-04T20:02:53Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) Nowadays, many High Performance Computing (HPC) centers with thousands of cores are available. The interconnection and synchronization between them is one of objectives of Grid Computing, as all these centers tend to form a global shared computing facility. In such scenario, dynamics of resource information is increasingly complex, and may become almost impossible to be captured using a static hierarchy. One of the key features needed in Grid Computing is to maintain up-to-date information about resources and services. Two important parameters that must be selected are the information refresh and expiration times. In this work, R is presented for studying the relation between these important parameters in three policies: Hierarchical, Super-peer and Random. The results reveal an interesting relation between expiration and refresh times in all studied policies. The study of this new parameter is essential to provide a deeper knowledge in new information distribution policies and in the configuration or deployment of existing ones.

2011-01-01T00:00:00Z Nowadays, many High Performance Computing (HPC) centers with thousands of cores are available. The interconnection and synchronization between them is one of objectives of Grid Computing, as all these centers tend to form a global shared computing facility. In such scenario, dynamics of resource information is increasingly complex, and may become almost impossible to be captured using a static hierarchy. One of the key features needed in Grid Computing is to maintain up-to-date information about resources and services. Two important parameters that must be selected are the information refresh and expiration times. In this work, R is presented for studying the relation between these important parameters in three policies: Hierarchical, Super-peer and Random. The results reveal an interesting relation between expiration and refresh times in all studied policies. The study of this new parameter is essential to provide a deeper knowledge in new information distribution policies and in the configuration or deployment of existing ones. Guzmán, Mateo Cabral, Matias Martínez Spessot, Cesar Ignacio http://sedici.unlp.edu.ar:80/handle/10915/126143 2021-10-04T20:03:03Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) Intel® Cluster Checker (CLCK) is a tool that verifies the configuration, measures performance and validates compliance of the Intel® Cluster Ready Specification in Linux-based clusters. With more than 100 tests-modules CLCK performs a wide range of cluster evaluations. It assesses firmware, kernel, storage, and network settings and also stresses compute nodes and network using multiple benchmarks. Customized checks can be created and integrated to the execution of the tool though an API provided for that purpose. However, this test modules should be written in Perl* programming language. This work proposes a new component of CLCK called Intel® Cluster Checker Multilanguage API (CLCK_Multilang_API), which allows users and developers to write their tailored test modules in arbitrary programming languages and add them to the execution of the tool. Two example languages, Perl* and Python were used as references for this project. The solution includes some mixing of technologies such as SWIG, native C APIs of each language, and Java Script Object Notation (JSON).

2011-01-01T00:00:00Z Intel® Cluster Checker (CLCK) is a tool that verifies the configuration, measures performance and validates compliance of the Intel® Cluster Ready Specification in Linux-based clusters. With more than 100 tests-modules CLCK performs a wide range of cluster evaluations. It assesses firmware, kernel, storage, and network settings and also stresses compute nodes and network using multiple benchmarks. Customized checks can be created and integrated to the execution of the tool though an API provided for that purpose. However, this test modules should be written in Perl* programming language. This work proposes a new component of CLCK called Intel® Cluster Checker Multilanguage API (CLCK_Multilang_API), which allows users and developers to write their tailored test modules in arbitrary programming languages and add them to the execution of the tool. Two example languages, Perl* and Python were used as references for this project. The solution includes some mixing of technologies such as SWIG, native C APIs of each language, and Java Script Object Notation (JSON). Monge, David A. Bělohradský, Jiří García Garino, Carlos Železný, Filip http://sedici.unlp.edu.ar:80/handle/10915/126141 2021-10-04T20:03:08Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) Through the years, scientific applications have demanded more powerful and sophisticated computing environments and management techniques. Workflows facilitated the design and management of scientific applications. The complexity of to day's workflows demand a high amount of resources and mechanisms for provisioning them. The execution of scientific workflow applications is a complex task and depends on how the resources are assigned. Scheduling is the name given to the process that assigns computing resources to the tasks comprised in a workflow. This work presents a scheduling algorithm (PPSA) for workflows tightly coupled to a performance prediction module (PEM). A set of experiments was developed for measuring the performance of the algorithm using the information provided by the proposed performance module. The proposed algorithm is compared with an algorithm included in the well-known workflow middlewares Condor DAGMan and ASKALON.

2011-01-01T00:00:00Z Through the years, scientific applications have demanded more powerful and sophisticated computing environments and management techniques. Workflows facilitated the design and management of scientific applications. The complexity of to day's workflows demand a high amount of resources and mechanisms for provisioning them. The execution of scientific workflow applications is a complex task and depends on how the resources are assigned. Scheduling is the name given to the process that assigns computing resources to the tasks comprised in a workflow. This work presents a scheduling algorithm (PPSA) for workflows tightly coupled to a performance prediction module (PEM). A set of experiments was developed for measuring the performance of the algorithm using the information provided by the proposed performance module. The proposed algorithm is compared with an algorithm included in the well-known workflow middlewares Condor DAGMan and ASKALON. Perez, Cristian Piccoli, Fabiana http://sedici.unlp.edu.ar:80/handle/10915/126135 2021-10-04T20:03:12Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) The characteristics of graphics processing units (GPUs), especially their parallel execution capabilities and fast memory access, render them attractive in many application areas. They promise more than an order of magnitude speedup over conventional processors for some non-graphics computations. The use of GPUs in general-purpose computing is becoming a very accepted alternative. In addition, the CUDA programming model gains acceptance. Each of these arguments make necessary count with tools that allow to evaluate GPUs. The performance parameters allow to model an architecture to predict the execution time of any application with any parallelism level. Furthermore, they are a useful tool to compare different architectures and determine its advantages and troubles. This work presents suitable benchmarks to evaluate different performance parameters of GPUs. The presented measurements focus on two issues of GPU performance: computing power and the global memory bandwidth. Their estimation will allow us determine technical characteristics of GPU and, in consequence, the analysis and optimization of the performance of applications that could run on actual or future GPUs.

2011-01-01T00:00:00Z The characteristics of graphics processing units (GPUs), especially their parallel execution capabilities and fast memory access, render them attractive in many application areas. They promise more than an order of magnitude speedup over conventional processors for some non-graphics computations. The use of GPUs in general-purpose computing is becoming a very accepted alternative. In addition, the CUDA programming model gains acceptance. Each of these arguments make necessary count with tools that allow to evaluate GPUs. The performance parameters allow to model an architecture to predict the execution time of any application with any parallelism level. Furthermore, they are a useful tool to compare different architectures and determine its advantages and troubles. This work presents suitable benchmarks to evaluate different performance parameters of GPUs. The presented measurements focus on two issues of GPU performance: computing power and the global memory bandwidth. Their estimation will allow us determine technical characteristics of GPU and, in consequence, the analysis and optimization of the performance of applications that could run on actual or future GPUs. Salinas, Sergio Ariel García Garino, Carlos Zunino, Alejandro http://sedici.unlp.edu.ar:80/handle/10915/126131 2021-10-04T20:03:17Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) La computación voluntaria permite la creación de Desktop Grids mediante la integración de distintos recursos dispersos en diferentes organizaciones. El objetivo de estos entornos es utilizar el tiempo ocioso de los recursos para ejecutar procesos de usuarios de la Grid. Los recursos en este tipo de Grid, se caracterizan por presentar un patrón de comportamiento que permite aplicar técnicas para predecir el estado de los mismos. De esta manera para cada instante de tiempo es posible predecir el estado del recurso. El sistema de planificación puede utilizar esta información para seleccionar aquellos recursos que mejor se ajusten a los requerimientos de tiempo de los procesos que debe asignar. Para ello es necesario proveer una heurística que permita comparar la predicción de disponibilidad de cada recurso. Esta información permite optimizar la asignación de procesos que realiza el sistema de planificación de la Grid Computational. En este trabajo, se presenta un sistema para predecir y evaluar la disponibilidad operativa de recursos en Desktop Grids. Los resultados muestran que el sistema de planificación obtiene mejoras en el tiempo de ejecución de procesos al utilizar la información provista por el sistema propuesto.

2011-01-01T00:00:00Z La computación voluntaria permite la creación de Desktop Grids mediante la integración de distintos recursos dispersos en diferentes organizaciones. El objetivo de estos entornos es utilizar el tiempo ocioso de los recursos para ejecutar procesos de usuarios de la Grid. Los recursos en este tipo de Grid, se caracterizan por presentar un patrón de comportamiento que permite aplicar técnicas para predecir el estado de los mismos. De esta manera para cada instante de tiempo es posible predecir el estado del recurso. El sistema de planificación puede utilizar esta información para seleccionar aquellos recursos que mejor se ajusten a los requerimientos de tiempo de los procesos que debe asignar. Para ello es necesario proveer una heurística que permita comparar la predicción de disponibilidad de cada recurso. Esta información permite optimizar la asignación de procesos que realiza el sistema de planificación de la Grid Computational. En este trabajo, se presenta un sistema para predecir y evaluar la disponibilidad operativa de recursos en Desktop Grids. Los resultados muestran que el sistema de planificación obtiene mejoras en el tiempo de ejecución de procesos al utilizar la información provista por el sistema propuesto. Tissera, Pablo Cristian Printista, Alicia Marcela Errecalde, Marcelo Luis http://sedici.unlp.edu.ar:80/handle/10915/126129 2021-10-04T20:03:23Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) Computer simulations using Cellular Automata (CA) have been applied with considerable success in different scientific areas, such as chemistry, biochemistry, economy, physics, etc. In this work we use CA in order to specify and implement a simulation model that allows to investigate behavioural dynamics for pedestrians in an emergency evacuation. Two important aspects must be considered when simulating the movement of people: a) estimation of distances from the cells to an exit and b) handling of collisions between individuals. For the first problem, the Dijkstra algorithm was used. In relation to the collisions, we proposed two approaches to solve the movement of people: centralised on a empty cell and distributed in the neighbouring cells. This latter approach leads to the formulation of Agent-based CA Model for pedestrians motion. Finally, in order to accelerate the simulation and take advantage of modern computer architectures, the paper also presents a parallel implementation which is an adaptation of the traditional Ghost Cell Pattern technique. This implementation will be essential when the model complexity increases due to the incorporation of new features. We apply our approaches to several environment configurations achieving important reduction of simulation time.

2011-01-01T00:00:00Z Computer simulations using Cellular Automata (CA) have been applied with considerable success in different scientific areas, such as chemistry, biochemistry, economy, physics, etc. In this work we use CA in order to specify and implement a simulation model that allows to investigate behavioural dynamics for pedestrians in an emergency evacuation. Two important aspects must be considered when simulating the movement of people: a) estimation of distances from the cells to an exit and b) handling of collisions between individuals. For the first problem, the Dijkstra algorithm was used. In relation to the collisions, we proposed two approaches to solve the movement of people: centralised on a empty cell and distributed in the neighbouring cells. This latter approach leads to the formulation of Agent-based CA Model for pedestrians motion. Finally, in order to accelerate the simulation and take advantage of modern computer architectures, the paper also presents a parallel implementation which is an adaptation of the traditional Ghost Cell Pattern technique. This implementation will be essential when the model complexity increases due to the incorporation of new features. We apply our approaches to several environment configurations achieving important reduction of simulation time. Printista, Alicia Marcela Saez, Fernando http://sedici.unlp.edu.ar:80/handle/10915/126126 2021-10-04T20:03:30Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) Cellular automata provide an abstract model of parallel computation that can be effectively used for modeling and simulation of complex phenomena and systems. We start from a template designed to facilitate faster D-dimensional cellular automata application development. The key for the use of the template is to achieve an efficient implementation, irrespective of the application specific details. In the parallel implementation on a cluster was important to consider issues such as task and data decomposition. With multicore clusters, new problems have emerged. The increasing numbers of cores per node, caches and shared memory inside the nodes, has led to the formation of a new hierarchy of access to processors. In this work we discuss and evaluate strategies that will be important in optimizing prototype to run on multicore cluster. The underlying idea in our proposal is the establishment of a relation among parallel processes based on the communication topology that arises in the implementation of task division functions. We propose that this relation can efficiently map on the multicore cluster topology. We introduce a new mapping strategy that can obtain benefit in the performance by adapting its communication pattern to the hardware affinities among processes allocated in different cores. We apply our approach to a two-dimensional application achieving sensible execution time reduction.

2011-01-01T00:00:00Z Cellular automata provide an abstract model of parallel computation that can be effectively used for modeling and simulation of complex phenomena and systems. We start from a template designed to facilitate faster D-dimensional cellular automata application development. The key for the use of the template is to achieve an efficient implementation, irrespective of the application specific details. In the parallel implementation on a cluster was important to consider issues such as task and data decomposition. With multicore clusters, new problems have emerged. The increasing numbers of cores per node, caches and shared memory inside the nodes, has led to the formation of a new hierarchy of access to processors. In this work we discuss and evaluate strategies that will be important in optimizing prototype to run on multicore cluster. The underlying idea in our proposal is the establishment of a relation among parallel processes based on the communication topology that arises in the implementation of task division functions. We propose that this relation can efficiently map on the multicore cluster topology. We introduce a new mapping strategy that can obtain benefit in the performance by adapting its communication pattern to the hardware affinities among processes allocated in different cores. We apply our approach to a two-dimensional application achieving sensible execution time reduction. Bederián, Carlos S. Dente, Axel D. http://sedici.unlp.edu.ar:80/handle/10915/126124 2021-10-04T20:03:33Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) The evolution calculation of quantum systems represents a great challenge nowadays. Numerical implementations typically scale exponentially with the size of the system, demanding high amounts of resources. General Purpose Graphics Processor Units (GPGPUs) enable a new range of possibilities for numerical simulations of quantum systems. In this work we implemented, optimized and compared the quantum Trotter-Suzuki algorithm running on both CPUs and GPUs.

2011-01-01T00:00:00Z The evolution calculation of quantum systems represents a great challenge nowadays. Numerical implementations typically scale exponentially with the size of the system, demanding high amounts of resources. General Purpose Graphics Processor Units (GPGPUs) enable a new range of possibilities for numerical simulations of quantum systems. In this work we implemented, optimized and compared the quantum Trotter-Suzuki algorithm running on both CPUs and GPUs. García, Sebastián Iturriaga, Santiago Nesmachnow, Sergio http://sedici.unlp.edu.ar:80/handle/10915/126123 2021-10-04T20:03:35Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) This work presents the application of parallel computing techniques to solve two scientific computing applications over the Latin America-Europe GISELA grid computing platform. The article describes two scientific computing applications –the semi-automatic processing of historical climate images and a software package for fluid dynamics– which usually require large computing times when applied to realistic scenarios. The proposal of applying parallel computing techniques over the GISELA grid infrastructure is formulated, and the implemented solutions are described. A preliminary experimental analysis is reported, presenting the estimated efficiency gains when using the grid infrastructure.

2011-01-01T00:00:00Z This work presents the application of parallel computing techniques to solve two scientific computing applications over the Latin America-Europe GISELA grid computing platform. The article describes two scientific computing applications –the semi-automatic processing of historical climate images and a software package for fluid dynamics– which usually require large computing times when applied to realistic scenarios. The proposal of applying parallel computing techniques over the GISELA grid infrastructure is formulated, and the implemented solutions are described. A preliminary experimental analysis is reported, presenting the estimated efficiency gains when using the grid infrastructure. Ravera, Emanuel Mariano Ferrer, Facundo http://sedici.unlp.edu.ar:80/handle/10915/126122 2021-10-04T20:03:41Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) Diskless clusters provide a stable, easily replicable and auto- deployable solution for High Performance Computing (HPC). On the other hand, the Intel® Cluster Ready program simplifies the selection, deployment, and management of HPC clusters by establishing a standard specification for them ensuring component and application interoperability. This work proposes to use a framework for on-the-fly creation of diskless solutions that allows users with little or no knowledge to create and deploy customized HPC clusters based on the Intel® Cluster Ready Specification.

2011-01-01T00:00:00Z Diskless clusters provide a stable, easily replicable and auto- deployable solution for High Performance Computing (HPC). On the other hand, the Intel® Cluster Ready program simplifies the selection, deployment, and management of HPC clusters by establishing a standard specification for them ensuring component and application interoperability. This work proposes to use a framework for on-the-fly creation of diskless solutions that allows users with little or no knowledge to create and deploy customized HPC clusters based on the Intel® Cluster Ready Specification. Ezzatti, Pablo Quintana-Ortí, Enrique S. Remón, Alfredo http://sedici.unlp.edu.ar:80/handle/10915/126121 2021-10-04T20:03:45Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) The solution of Algebraic Riccati Equations is required in many linear optimal and robust control methods such as LQR, LQG, Kalman filter, and in model order reduction techniques like the balanced stochastic truncation method. Numerically reliable algorithms for these applications rely on the sign function method, and require O(8n3) floating-point arithmetic operations, with n in the range of 103 −105 for many practical applications. In this paper we investigate the use of graphics processors (GPUs) to accelerate the solution of Algebraic Riccati Equations by off-loading the computationally intensive kernels to this device. Experiments on a hybrid platform compose by state-of-the-art general-purpose multi-core processors and a GPU illustrate the potential of this approach.

2011-01-01T00:00:00Z The solution of Algebraic Riccati Equations is required in many linear optimal and robust control methods such as LQR, LQG, Kalman filter, and in model order reduction techniques like the balanced stochastic truncation method. Numerically reliable algorithms for these applications rely on the sign function method, and require O(8n3) floating-point arithmetic operations, with n in the range of 103 −105 for many practical applications. In this paper we investigate the use of graphics processors (GPUs) to accelerate the solution of Algebraic Riccati Equations by off-loading the computationally intensive kernels to this device. Experiments on a hybrid platform compose by state-of-the-art general-purpose multi-core processors and a GPU illustrate the potential of this approach. Carballal, Claudio A. Hamkalo, José Luis Cernuschi-Frías, Bruno http://sedici.unlp.edu.ar:80/handle/10915/126119 2021-10-04T20:03:52Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) This work presents a study of fairness in cache sharing between processes in a chip multiprocessor (CMP). We propose a new algorithm that uses a metric based on the D3C miss classification and LRU Stack Distance, to measure the fairness in the administration of the resources to achieve an increase of the global IPC of all executed processes. Shared cache miss rate, IPC and bandwidth metrics were considered to analyze the simulation results obtained using three test sets. The obtained results showed that the proposed dynamic management policy compared to Capitalist management policy, has a lower global miss rate in shared cache and lower bandwidth usage for each test set studied and fulfills its objective of managing the shared cache space for every process while improving the overall IPC.

2011-01-01T00:00:00Z This work presents a study of fairness in cache sharing between processes in a chip multiprocessor (CMP). We propose a new algorithm that uses a metric based on the D3C miss classification and LRU Stack Distance, to measure the fairness in the administration of the resources to achieve an increase of the global IPC of all executed processes. Shared cache miss rate, IPC and bandwidth metrics were considered to analyze the simulation results obtained using three test sets. The obtained results showed that the proposed dynamic management policy compared to Capitalist management policy, has a lower global miss rate in shared cache and lower bandwidth usage for each test set studied and fulfills its objective of managing the shared cache space for every process while improving the overall IPC. Vázquez, M. Lafortune, P. Houzeaux, G. Arís, R. http://sedici.unlp.edu.ar:80/handle/10915/126116 2021-10-04T20:03:54Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) In this paper, we present a high performance computational electromechanical model of the heart, coupling between electrical activation and mechanical deformation and running efficiently in up to thousands of processors. The electrical potential propagation is modelled by FitzHugh-Nagumo or Fenton-Karma models, with fiber orientation. The mechanical deformation is treated using anisotropic hyper-elastic materials in a total Lagrangian formulation. Several material models are assessed, such as models based on biaxial tests on excised myocardium or orthotropic formulations. Coupling is treated using the Cross-Bridges model of Peterson. The scheme is implemented in Alya, which run simulations in parallel with almost linear scalability in a wide range computer sizes, up to thousands of processors. The computational model is assessed through several tests, including those to evaluate its parallel performance.

2011-01-01T00:00:00Z In this paper, we present a high performance computational electromechanical model of the heart, coupling between electrical activation and mechanical deformation and running efficiently in up to thousands of processors. The electrical potential propagation is modelled by FitzHugh-Nagumo or Fenton-Karma models, with fiber orientation. The mechanical deformation is treated using anisotropic hyper-elastic materials in a total Lagrangian formulation. Several material models are assessed, such as models based on biaxial tests on excised myocardium or orthotropic formulations. Coupling is treated using the Cross-Bridges model of Peterson. The scheme is implemented in Alya, which run simulations in parallel with almost linear scalability in a wide range computer sizes, up to thousands of processors. The computational model is assessed through several tests, including those to evaluate its parallel performance. Caymes-Scutari, Paola Bianchini, Germán http://sedici.unlp.edu.ar:80/handle/10915/126115 2021-10-04T20:03:56Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) The use of high performance computing has been gaining more and more followers in the different branches of Science and Engineering given the potential offered to deal with complex and big problems, and the increasing economic facilities to configure some kind of parallel machine. However, the parallel programming paradigm involves additional aspects to the merely functional which could provoke different kinds of bottlenecks in the performance of the applications. Such difficulties may represent critical obstacles specially for the non-expert users. In this paper we present an environment to provide general support for the automatic development and tuning of parallel applications. The environment provides an interface to guide the user in the specification of the problem and the solution, which makes transparent the process of code generation and instrumentation. Because the parallelization of any problem/solution is hard to generalize, the environment tackles different classes of problems. In this article, we introduce the class of Parallel Genetic Algorithms.

2011-01-01T00:00:00Z The use of high performance computing has been gaining more and more followers in the different branches of Science and Engineering given the potential offered to deal with complex and big problems, and the increasing economic facilities to configure some kind of parallel machine. However, the parallel programming paradigm involves additional aspects to the merely functional which could provoke different kinds of bottlenecks in the performance of the applications. Such difficulties may represent critical obstacles specially for the non-expert users. In this paper we present an environment to provide general support for the automatic development and tuning of parallel applications. The environment provides an interface to guide the user in the specification of the problem and the solution, which makes transparent the process of code generation and instrumentation. Because the parallelization of any problem/solution is hard to generalize, the environment tackles different classes of problems. In this article, we introduce the class of Parallel Genetic Algorithms. Raimondo, Federico Kamienkowski, Juan E. Fernández Slezak, Diego http://sedici.unlp.edu.ar:80/handle/10915/126113 2021-10-04T20:03:59Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) Nowadays, with the advent of new non-invasive techniques of brain imaging, researchers have access to neural processes underlying the cognition in humans. One of the main challenges in this techniques is the detection of patterns in brain signals, generally very noisy and with artifacts inserted by vital signs. One of the most successful techniques for this is Independent Component Analysis which detects statistically independent components that are produced from different sources. These methods are very expensive in computational time, with many hours of processing for a single experiment. We analyzed this algorithm and detect two main types of operations: vector-matrix and matrix-matrix. We implemented an ad-hoc solution that executes on GPU and compared this with the original and CUBLAS versions. We obtained a 4x and 40x of performance increase of vector-matrix and matrix-matrix operations, respectively. These results are the first step towards real-time EEG processing which may produce a significant advance into BCI applications.

2011-01-01T00:00:00Z Nowadays, with the advent of new non-invasive techniques of brain imaging, researchers have access to neural processes underlying the cognition in humans. One of the main challenges in this techniques is the detection of patterns in brain signals, generally very noisy and with artifacts inserted by vital signs. One of the most successful techniques for this is Independent Component Analysis which detects statistically independent components that are produced from different sources. These methods are very expensive in computational time, with many hours of processing for a single experiment. We analyzed this algorithm and detect two main types of operations: vector-matrix and matrix-matrix. We implemented an ad-hoc solution that executes on GPU and compared this with the original and CUBLAS versions. We obtained a 4x and 40x of performance increase of vector-matrix and matrix-matrix operations, respectively. These results are the first step towards real-time EEG processing which may produce a significant advance into BCI applications. Bianchini, Germán Caymes-Scutari, Paola http://sedici.unlp.edu.ar:80/handle/10915/126112 2021-10-04T20:04:02Z 2011-01-01T00:00:00Z

Objeto de conferencia IV High-Performance Computing Symposium (HPC 2011) (XL JAIIO, Córdoba, 31 de agosto y 1º de septiembre de 2011) In many scientific areas, the use of models to represent physical systems has become a common strategy. These models receive some input parameters representing some particular conditions and they provide an output representing the evolution of the system. Usually, these models are integrated into simulation tools that can be executed on a computational system. A particular case where models are very useful is the prediction of Forest Fire propagation. Therefore, the use of models is very relevant to estimate fire risk and to predict fire behaviour. However, in many cases the models present a series of limitations. Such restrictions are due to the need for a large number of input parameters and, usually, such parameters present some uncertainty due to the impossibility of measuring all of them in real time. In consequence, they have to be estimated from indirect measurements. To overcome this drawback and improve the quality of the prediction, in this work we propose a method that combines Statistical Analysis and Parallel Evolutionary Algorithms.

2011-01-01T00:00:00Z In many scientific areas, the use of models to represent physical systems has become a common strategy. These models receive some input parameters representing some particular conditions and they provide an output representing the evolution of the system. Usually, these models are integrated into simulation tools that can be executed on a computational system. A particular case where models are very useful is the prediction of Forest Fire propagation. Therefore, the use of models is very relevant to estimate fire risk and to predict fire behaviour. However, in many cases the models present a series of limitations. Such restrictions are due to the need for a large number of input parameters and, usually, such parameters present some uncertainty due to the impossibility of measuring all of them in real time. In consequence, they have to be estimated from indirect measurements. To overcome this drawback and improve the quality of the prediction, in this work we propose a method that combines Statistical Analysis and Parallel Evolutionary Algorithms.