Juan José Escobar

Parallelization of the French Meteorological Mesoscale Model MésoNH

Numerical simulation of the atmospheric motions requires the most powerful machines and a high performance computer technology. The French meteorological model MesoNH is designed as a research tool for small and mesoscale atmospheric processes. This paper describes softwars and techniques used for implementing this numerical model on parallel processor computers and presents first results and performances.

Assessing Parallel Heterogeneous Computer Architectures for Multiobjective Feature Selection on EEG Classification

High-dimensional multi-objective optimization will open promising approaches to many applications on bioinformatics once efficient parallel procedures are available. These procedures have to take advantage of the present heterogeneous architectures comprising multicore CPUs and GPUs. In this paper, we describe and analyze several OpenCL implementations for an application comprising multiobjective feature selection for clustering in an EEG classification task on high-dimensional patterns. These implementation alternatives correspond to different uses of multicore CPU and GPU platforms to process irregular data codes. Depending on the dataset used, we have reached speedups of up to 14.9 and 17.2 with up to 24 threads for the implemented OpenCL CPU kernels and of up to 7.1 and 9.1 with up to 13 SMX processors and 256 local work-items for our OpenCL GPU kernels. Nevertheless, to provide this level of performance, careful considerations about the use of the memory hierarchy of the heterogeneous architecture and different strategies to cope with the irregularity of our target application have to be taken into account.

Improving Memory Accesses for Heterogeneous Parallel Multi-objective Feature Selection on EEG Classification

Bioinformatics applications that analyze large volumes of high-dimensional data and present different implicit parallelism can benefit from the efficient use, in performance terms, of heterogeneous parallel architectures, including accelerators such as graphics processing units (GPUs). This paper aims to take advantage of parallel codes to accelerate electroencephalogram (EEG) classification and feature selection problems in the context of Branch-Computing Interface (BCI) tasks. As the approaches to tackle these applications usually involve optimized codes that implement different types of parallelism, the use of heterogeneous architectures with multicore microprocessors along with GPUs could provide relevant performance improvements after careful code optimizing. More specifically, the memory access patterns have been taken into account to improve the performance of data-parallel GPU kernels.

Parallel high-dimensional multi-objective feature selection for EEG classification with dynamic workload balancing on CPU–GPU architectures

Many bioinformatics applications that analyse large volumes of high-dimensional data comprise complex problems requiring metaheuristics approaches with different types of implicit parallelism. For example, although functional parallelism would be used to accelerate evolutionary algorithms, the fitness evaluation of the population could imply the computation of cost functions with data parallelism. This way, heterogeneous parallel architectures, including central processing unit (CPU) microprocessors with multiple superscalar cores and accelerators such as graphics processing units (GPUs) could be very useful. This paper aims to take advantage of such CPU–GPU heterogeneous architectures to accelerate electroencephalogram classification and feature selection problems by evolutionary multi-objective optimization, in the context of brain computing interface tasks. In this paper, we have used the OpenCL framework to develop parallel master-worker codes implementing an evolutionary multi-objective feature selection procedure in which the individuals of the population are dynamically distributed among the available CPU and GPU cores.

Power-Performance Evaluation of Parallel Multi-objective EEG Feature Selection on CPU-GPU Platforms

Heterogeneous CPU-GPU platforms include resources to benefit from different kinds of parallelism present in many data mining applications based on evolutionary algorithms that evolve solutions with time-demanding fitness evaluation. This paper describes an evolutionary parallel multi-objective feature selection procedure with subpopulations using two scheduling alternatives for evaluation of individuals according to the number of subpopulations. Evolving subpopulations usually provides good diversity properties and avoids premature convergence in evolutionary algorithms. The proposed procedure has been implemented in OpenMP to distribute dynamically either subpopulations or individuals among devices and OpenCL to evaluate the individuals taking into account the devices characteristics, providing two parallelism levels in CPU and up to three levels in GPUs. Different configurations of the proposed procedure have been evaluated and compared with a master-worker approach considering not only the runtime and achieved speedups but also the energy consumption between both scheduling models.

Multi-objective feature selection for EEG classification with multi-level parallelism on heterogeneous CPU-GPU clusters

The present trend in the development of computer architectures that offer improvements in both performance and energy efficiency has provided clusters with interconnected nodes including multiple multi-core microprocessors and accelerators. In these so-called heterogeneous computers, the applications can take advantage of different parallelism levels according to the characteristics of the architectures in the platform. Thus, the applications should be properly programmed to reach good efficiencies, not only with respect to the achieved speedups but also taking into account the issues related to energy consumption. In this paper we provide a multi-objective evolutionary algorithm for feature selection in electroencephalogram (EEG) classification, which can take advantage of parallelism at multiple levels: among the CPU-GPU nodes interconnected in the cluster (through message-passing), and inside these nodes (through shared-memory thread-level parallelism in the CPU cores, and data-level parallelism and thread-level parallelism in the GPU). The procedure has been experimentally evaluated in performance and energy consumption and shows statistically significant benefits for feature selection: speedups of up to 73 requiring only a 6% of the energy consumed by the sequential code.

Energy-aware load balancing of parallel evolutionary algorithms with heavy fitness functions in heterogeneous CPU-GPU architectures: Energy-aware load balancing of parallel evolutionary algorithms with heavy fitness functions in heterogeneous CPU-GPU architectures

By means of the availability of mechanisms such as Dynamic Voltage and Frequency Scaling (DVFS) and heterogeneous architectures including processors with different power consumption profiles, it is possible to devise scheduling algorithms that are aware of both runtime and energy consumption in parallel programs. In this paper, we propose and evaluate a multi‐objective (more specifically, a bi‐objective) approach to distribute the workload among the processing cores in a given heterogeneous parallel CPU‐GPU architecture. The aim of this distribution may be either to save energy without increasing the running time or to reach a trade‐off among time and energy consumption. The parallel programs considered here are master‐worker evolutionary algorithms where the evaluation of the fitness function for the individuals in the population demands the most part of the computing time. As many useful bioinformatics and data mining applications exhibit this kind of parallel profile, the proposed energy‐aware approach for workload scheduling could be frequently applied.

Issues on GPU Parallel Implementation of Evolutionary High-Dimensional Multi-objective Feature Selection

The interest on applications that analyse large volumes of high dimensional data has grown recently. Many of these applications related to the so-called Big Data show different implicit parallelism that can benefit from the efficient use, in terms of performance and power consumption, of Graphics Processing Unit (GPU) accelerators. Although the GPU microarchitectures make possible the acceleration of applications by exploiting parallelism at different levels, the characteristics of their memory hierarchy and the location of GPUs as coprocessors require a careful organization of the memory access patterns and data transferences to get efficient speedups. This paper aims to take advantage of heterogeneous parallel codes on GPUs to accelerate evolutionary approaches in Electroencephalogram (EEG) classification and feature selection in the context of Brain Computer Interface (BCI) tasks. The results show the benefits of taking into account not only the data parallelism achievable by GPUs, but also the memory access patterns, in order to increase the speedups achieved by superscalar cores.

Speedup and Energy Analysis of EEG Classification for BCI Tasks on CPU-GPU Clusters

Many data mining applications on bioinformatics and bioengineering require solving problems with different profiles from the point of view of their implicit parallelism. In this context, heterogeneous architectures comprised by interconnected nodes with multiple multi-core microprocessors and accelerators, such as vector processors, Graphics Processing Units (GPUs), or Field-Programmable Gate Arrays would constitute suitable platforms that offer the possibility of not only to accelerate the running time of the applications, but also to optimize the energy consumption. In this paper, we analyze the speedups and energy consumption of a parallel multiobjective approach for feature selection and classification of electroencephalograms in Brain Computing Interface tasks, by considering different implementation alternatives in a heterogeneous CPU-GPU cluster. The procedure is able to take advantage of parallelism through message-passing among the CPU-GPU nodes of the cluster (through shared-memory and thread-level parallelism in the CPU cores, and data-level parallelism and thread-level parallelism in the GPU). The experimental results show high code accelerations and high energy-savings: running times between 1.4 and 5.3% of the sequential time and energy consumptions between 5.9 and 11.6% of the energy consumed by the sequential execution.