High Performance Computing Techniques Research Articles

Moving from exascale to zettascale computing: challenges and techniques

High-performance computing (HPC) is essential for both traditional and emerging scientific fields, enabling scientific activities to make progress. With the development of high-performance computing, it is foreseeable that exascale computing will be put into practice around 2020. As Moore’s law approaches its limit, high-performance computing will face severe challenges when moving from exascale to zettascale, making the next 10 years after 2020 a vital period to develop key HPC techniques. In this study, we discuss the challenges of enabling zettascale computing with respect to both hardware and software. We then present a perspective of future HPC technology evolution and revolution, leading to our main recommendations in support of zettascale computing in the coming future.

DYNAMICAL PREDICTION TECHNIQUE FOR GEOSIMULATION USING MULTISPECTRAL REMOTE SENSING DATA

Abstract. The analysis of dynamical models for prediction and geosimulation using the information extracted from a geographical region processed from the data provided by multispectral remote sensing systems provides useful information for urban planning and resource management. However, several topics of interest on this particular matter are still to be properly studied. Using the remote sensing data that has been extracted from multispectral images from a particular geographic region in discrete time, its dynamic study is performed in both, spatial resolution and time evolution, in order to obtain the dynamical model of the physical variables and the evolutionary information about the data. This provides a background for understanding the future trends in development of the dynamics inherent in the multispectral and high-resolution images. This proposition is performed via an intelligent computational paradigm based on the use of dynamical filtering techniques modified to enhance the quality of reconstruction of the data extracted from multispectral remote sensing images and using high-performance computational techniques to unify the available data scheme with its dynamic analysis and, therefore, provide a behavioral model of the sensed data.

Vlasov methods in space physics and astrophysics

This paper reviews Vlasov-based numerical methods used to model plasma in space physics and astrophysics. Plasma consists of collectively behaving charged particles that form the major part of baryonic matter in the Universe. Many concepts ranging from our own planetary environment to the Solar system and beyond can be understood in terms of kinetic plasma physics, represented by the Vlasov equation. We introduce the physical basis for the Vlasov system, and then outline the associated numerical methods that are typically used. A particular application of the Vlasov system is Vlasiator, the world’s first global hybrid-Vlasov simulation for the Earth’s magnetic domain, the magnetosphere. We introduce the design strategies for Vlasiator and outline its numerical concepts ranging from solvers to coupling schemes. We review Vlasiator’s parallelisation methods and introduce the used high-performance computing (HPC) techniques. A short review of verification, validation and physical results is included. The purpose of the paper is to present the Vlasov system and introduce an example implementation, and to illustrate that even with massive computational challenges, an accurate description of physics can be rewarding in itself and significantly advance our understanding. Upcoming supercomputing resources are making similar efforts feasible in other fields as well, making our design options relevant for others facing similar challenges.

Inpactor, Integrated and Parallel Analyzer and Classifier of LTR Retrotransposons and Its Application for Pineapple LTR Retrotransposons Diversity and Dynamics.

One particular class of Transposable Elements (TEs), called Long Terminal Repeats (LTRs), retrotransposons, comprises the most abundant mobile elements in plant genomes. Their copy number can vary from several hundreds to up to a few million copies per genome, deeply affecting genome organization and function. The detailed classification of LTR retrotransposons is an essential step to precisely understand their effect at the genome level, but remains challenging in large-sized genomes, requiring the use of optimized bioinformatics tools that can take advantage of supercomputers. Here, we propose a new tool: Inpactor, a parallel and scalable pipeline designed to classify LTR retrotransposons, to identify autonomous and non-autonomous elements, to perform RT-based phylogenetic trees and to analyze their insertion times using High Performance Computing (HPC) techniques. Inpactor was tested on the classification and annotation of LTR retrotransposons in pineapple, a recently-sequenced genome. The pineapple genome assembly comprises 44% of transposable elements, of which 23% were classified as LTR retrotransposons. Exceptionally, 16.4% of the pineapple genome assembly corresponded to only one lineage of the Gypsy superfamily: Del, suggesting that this particular lineage has undergone a significant increase in its copy numbers. As demonstrated for the pineapple genome, Inpactor provides comprehensive data of LTR retrotransposons’ classification and dynamics, allowing a fine understanding of their contribution to genome structure and evolution. Inpactor is available at https://github.com/simonorozcoarias/Inpactor.

Big Data and IT Network Data Visualization

Visualization with graphs is popular in the data analysis of Information Technology (IT) networks or computer networks. An IT network is often modelled as a graph with hosts being nodes and traffic being flows on many edges. General visualization methods are introduced in this paper. Applications and technology progress of visualization in IT network analysis and big data in IT network visualization are presented. The challenges of visualization and Big Data analytics in IT network visualization are also discussed. Big Data analytics with High Performance Computing (HPC) techniques, especially Graphics Processing Units (GPUs) helps accelerate IT network analysis and visualization.

Aiding Cascading Analysis Modelling with High-performance-computing Technology

Abstract Cascading outage analysis is important to minimize the risk of power system blackouts. To sustain power system reliability and manage the risk of blackouts requires simulation tools that are able to analyze and assess the consequence of cascading events quickly and efficiently. This paper presents a cascading analysis feature that was recently implemented upon an integrated state estimation and contingency analysis software package to utilize the power of high performance computing (HPC) techniques. The performance of the developed cascading analysis function, including the consequence analysis and scalability, is demonstrated through three test systems that represent small, medium, and large power systems. The package is able to perform cascading analysis with full N-1 contingency for a European open model within 20 seconds with 2,048 cores. A discussion of HPC implementation is also provided.

A Feature Selection Model based on High-Performance Computing (HPC) Techniques

A Feature Selection Model based on High-Performance Computing (HPC) Techniques

On Parallelizing Single Dynamic Simulation Using HPC Techniques and APIs of Commercial Software

Time-domain simulations are heavily used in today's planning and operation practices to assess power system transient stability, posttransient voltage/frequency profiles as well as oscillatory behavior following severe contingencies to comply with industry standards. Because of the increased modeling complexity, it is many times slower than real time for state-of-the-art commercial software packages to complete a dynamic simulation for a large-scale model. With the growing stochastic behavior introduced by emerging technologies, power industry has seen a growing need for performing security assessment in real time. This paper presents an innovative parallel implementation framework to speed up a single dynamic simulation by leveraging the existing stability model library in commercial tools through their application programming interfaces (APIs). Several high performance computing (HPC) techniques are explored and implemented such as parallelizing the calculation of generator current injection, identifying fast linear solvers for network solution, and parallelizing data outputs when interacting with APIs in the commercial package, TSAT. The proposed method has been tested on a Western Electricity Coordinating Council (WECC) planning base case with detailed synchronous generator models and it exhibits outstanding scalable performance with sufficient accuracy.

Editorial: Fourth Quarter 2017 IEEE Communications Surveys and Tutorials

I Welcome you to the fourth issue of the IEEE Communications Surveys and Tutorials in 2017. This issue includes 35 papers covering different aspects of communication networks. In particular, these articles survey and tutor various issues in “Wireless, Sensor, and Cognitive Radio Networks,” “Software-Defined Networking, Mobile Edge Computing, and Intelligent Networks,” “Internet of Things and Vehicular Networks,” “Security, High-Performance Computing, Positioning, and Localization Techniques,” and “Internet, Multimedia, Social Network, Molecular Network, and Optical Communication.” A brief account for each of these papers is given below.

Editorial: First Quarter 2017 IEEE Communications Surveys and Tutorials

I Welcome you to the fourth issue of the IEEE Communications Surveys and Tutorials in 2017. This issue includes 35 papers covering different aspects of communication networks. In particular, these articles survey and tutor various issues in “Wireless, Sensor, and Cognitive Radio Networks,” “Software-Defined Networking, Mobile Edge Computing, and Intelligent Networks,” “Internet of Things and Vehicular Networks,” “Security, High-Performance Computing, Positioning, and Localization Techniques,” and “Internet, Multimedia, Social Network, Molecular Network, and Optical Communication.” A brief account for each of these papers is given below.

Full Three-Dimensional Seismic Response Analysis of Underground Structures with Large Complex Cross Sections and Two-Step Analysis Method for Reducing the Computational Costs

To enhance the reliability of estimates of seismic behavior, with a special emphasis on quality assurance of numerical simulations, this paper presents a full three-dimensional (3D) seismic response analysis of a large underground structure with a complex cross section. We conduct a full 3D seismic-response analysis using a high-fidelity model with quality assurance and a high-performance computing technique in a supercomputer environment. Due to the large computational costs of such analyses, we propose a two-step numerical simulation method based on multi-scale analysis, image-based modeling, and sophisticated approximation techniques. We provide numerical examples showing that the method can successfully reproduce the dynamic behavior of a full 3D model, but with a considerable reduction in computational costs. Our method is expected to support practical seismic safety inspections and the design of underground structures with complex large-cross-sectional configurations, as it has already been applied to the practical seismic safety inspection of an actual underground highway.

Parallel Hybrid Method of HOMoM–MLFMA for Analysis of Large Antenna Arrays on an Electrically Large Platform

The higher order method of moments (HOMoM) with a direct solver is hybridized with the multilevel fast multipole algorithm given the multiscale nature of airborne antenna arrays. The Poggio–Miller–Chang–Harrington–Wu formulation and the combined filed integral equation are solved for modeling metallic/dielectric antennas on a conducting platform. To accelerate the outer iteration between the two algorithms, the lower/upper decomposed matrix in HOMoM is reused and the initial values of current coefficients are properly set. Furthermore, a computational scheme based on the high-performance computing technique is designed to improve the hybrid method. A large model of a 2132-element microstrip array mounted on a 1020-wavelength airplane is accurately simulated.

A hybrid MPI/OpenMP parallel implementation of NSGA-II for finding patterns in protein sequences

Since the late 1970s, when the first DNA-based genome was sequenced, the field of biology is experiencing a significant growth in the amount of data that needs to be processed. Long ago it became impractical to analyze all this information manually, resulting in a great need for new techniques, algorithms and strategies to facilitate this work. Within the vast world of bioinformatics, we will focus on proteomics, more specifically, on the discovery of small repeated common patterns on sets of protein sequences that may represent some biological functionality. When we analyze a large number of sequences, the problem shows non-deterministic polynomial times, it implies that we could benefit from the combination of high-performance computing and computational intelligence techniques. In this paper, we address the discovery of repeated common patterns as a multiobjective optimization problem by means of a hybrid MPI/OpenMP approach which parallelizes a well-known multiobjective metaheuristic, the fast non-dominated sorting genetic algorithm (NSGA-II). Our main objective is to combine the benefits of shared-memory and distributed-memory programming paradigms to discover patterns in an accurate and efficient manner. Experiments conducted on six different datasets, comparisons with other well-known biological tools, and the obtained speed-ups and efficiencies show that our approach is able to achieve a significant performance in terms of parallel and biological results.

Care HPS: A high performance simulation tool for parallel and distributed agent-based modeling

Parallel and distributed simulation is a powerful tool for developing complex agent-based simulation. Complex simulations require parallel and distributed high performance computing solutions. It is necessary because their sequential solutions are not able to give answers in a feasible total execution time. Therefore, for the advance of computing science, it is important that High Performance Computing (HPC) techniques and solutions be proposed and studied. In literature, we can find some agent-based modeling and simulation tools that use HPC. However, none of these tools are designed to enable the HPC expert to be able to propose new techniques and solutions without great effort. In this paper, we introduce Care High Performance Simulation (HPS), which is a scientific instrument that enables researchers to: (1) develop techniques and solutions of high performance distributed simulations for agent-based models; and, (2) study, design and implement complex agent-based models that require HPC solutions. Care HPS was designed to easily and quickly develop new agent-based models. It was also designed to extend and implement new solutions for the main issues of parallel and distributed solutions such as: synchronization, communication, load and computing balancing, and partitioning algorithms. We conducted some experiments with the aim of showing the completeness and functionality of Care HPS. As a result, we show that Care HPS can be used as a scientific instrument for the advance of the agent-based parallel and distributed simulations field.

Coarse grain parallelization of deep neural networks

Deep neural networks (DNN) have recently achieved extraordinary results in domains like computer vision and speech recognition. An essential element for this success has been the introduction of high performance computing (HPC) techniques in the critical step of training the neural network. This paper describes the implementation and analysis of a network-agnostic and convergence-invariant coarse-grain parallelization of the DNN training algorithm. The coarse-grain parallelization is achieved through the exploitation of the batch-level parallelism. This strategy is independent from the support of specialized and optimized libraries. Therefore, the optimization is immediately available for accelerating the DNN training. The proposal is compatible with multi-GPU execution without altering the algorithm convergence rate. The parallelization has been implemented in Caffe, a state-of-the-art DNN framework. The paper describes the code transformations for the parallelization and we also identify the limiting performance factors of the approach. We show competitive performance results for two state-of-the-art computer vision datasets, MNIST and CIFAR-10. In particular, on a 16-core Xeon E5-2667v2 at 3.30GHz we observe speedups of 8× over the sequential execution, at similar performance levels of those obtained by the GPU optimized Caffe version in a NVIDIA K40 GPU.

MultiGeMS: detection of SNVs from multiple samples using model selection on high-throughput sequencing data.

Single nucleotide variant (SNV) detection procedures are being utilized as never before to analyze the recent abundance of high-throughput DNA sequencing data, both on single and multiple sample datasets. Building on previously published work with the single sample SNV caller genotype model selection (GeMS), a multiple sample version of GeMS (MultiGeMS) is introduced. Unlike other popular multiple sample SNV callers, the MultiGeMS statistical model accounts for enzymatic substitution sequencing errors. It also addresses the multiple testing problem endemic to multiple sample SNV calling and utilizes high performance computing (HPC) techniques. A simulation study demonstrates that MultiGeMS ranks highest in precision among a selection of popular multiple sample SNV callers, while showing exceptional recall in calling common SNVs. Further, both simulation studies and real data analyses indicate that MultiGeMS is robust to low-quality data. We also demonstrate that accounting for enzymatic substitution sequencing errors not only improves SNV call precision at low mapping quality regions, but also improves recall at reference allele-dominated sites with high mapping quality. The MultiGeMS package can be downloaded from https://github.com/cui-lab/multigems xinping.cui@ucr.edu Supplementary data are available at Bioinformatics online.

Prefiltering Model for Homology Detection Algorithms on GPU.

Homology detection has evolved over the time from heavy algorithms based on dynamic programming approaches to lightweight alternatives based on different heuristic models. However, the main problem with these algorithms is that they use complex statistical models, which makes it difficult to achieve a relevant speedup and find exact matches with the original results. Thus, their acceleration is essential. The aim of this article was to prefilter a sequence database. To make this work, we have implemented a groundbreaking heuristic model based on NVIDIA’s graphics processing units (GPUs) and multicore processors. Depending on the sensitivity settings, this makes it possible to quickly reduce the sequence database by factors between 50% and 95%, while rejecting no significant sequences. Furthermore, this prefiltering application can be used together with multiple homology detection algorithms as a part of a next-generation sequencing system. Extensive performance and accuracy tests have been carried out in the Spanish National Centre for Biotechnology (NCB). The results show that GPU hardware can accelerate the execution times of former homology detection applications, such as National Centre for Biotechnology Information (NCBI), Basic Local Alignment Search Tool for Proteins (BLASTP), up to a factor of 4. KEY POINTS:Owing to the increasing size of the current sequence datasets, filtering approach and high-performance computing (HPC) techniques are the best solution to process all these information in acceptable processing times.Graphics processing unit cards and their corresponding programming models are good options to carry out these processing methods.Combination of filtration models with HPC techniques is able to offer new levels of performance and accuracy in homology detection algorithms such as National Centre for Biotechnology Information Basic Local Alignment Search Tool.

S-UTD-CH model in multiple diffractions

ABSTRACTThe propagation of electromagnetic waves in empty space is an extremely simplified case. Thus, the significant question is how an electromagnetic wave propagates in an environment with obstacles such as buildings, trees or hills. Electromagnetic waves are partially reflected and partially diffracted from these obstacles. To predict the relative path loss of electromagnetic waves at the receiving position, many electromagnetic-wave propagation models have been proposed. These propagation models can be classified into models based on numerical integration and those based on ray tracing. Uniform theory of diffraction (UTD) and slope-UTD (S-UTD) models are ray-tracing-based propagation models and are briefly explained in this paper. In addition, detailed information is provided about the improved slope UTD model, which is called the S-UTD with Convex Hull (S-UTD-CH) model. The fundamentals of the S-UTD-CH model are the S-UTD, convex hull and Fresnel zone concept. In particular, the S-UTD-CH model can be applied to multiple diffraction scenarios in the transition region. Moreover, the S-UTD-CH model is considered an optimum model in terms of its accuracy and calculation or computation time. Widespread simulation results are provided to compare the models based on theoretical rays in terms of prediction accuracy and computation time. To compare these models, different operation frequencies and transmitting antenna heights are considered by using a high-performance computing technique.

English

English

Image Processing Techniques for Contrast Enhancement with Poor Lighting on Social and Medical Images

enhancement is a technique that increases the visual contrast in a designated intensity range. Contrast is an act of distinguishing by comparing differences. Morphological transformation and block analysis are used to detect the background of various social and medical images. Opening by reconstruction method of contrast image transformation can be defined by two operators - opening and closing. The first operator makes use of the information from block analysis, while the second transformation utilizes the opening by reconstruction. The Later is used to define the multi background notion. The complete image processing is being implemented using JAVA simulation model. Quality of image enhancement is assessed by different techniques. In this paper, High performance Computational techniques involving contrast enhancement and noise filtering on various medical , social images are developed using Weber's law. Image quality assessment is compared by different techniques. The values of all the quality assessment parameters are found to be in the standard expected ranges thereby confirming the enhancement of quality of images. Keywordstransformation, morphological reconstruction,