High-performance Parallel Computing Research Articles

Developments and needs in nuclear analysis of fusion technology

Nuclear analyses provide essential input to the conceptual design, optimisation, engineering and safety case of fusion technology in current experiments, ITER, next-step devices and power plant studies. Calculations are intricate and computer-intensive, typically requiring detailed geometry models, sophisticated acceleration algorithms, high-performance parallel computations, and coupling of large and complex transport and activation codes and databases. This paper reports progress on some key areas in the development of tools and methods to meet the specific needs of fusion nuclear analyses. In particular, advances in the production and modernisation of reference models, in the preparation and quality assurance of acceleration algorithms and coupling schemes, and in the evaluation and adaptation of alternative transport codes are presented. Emphasis is given to ITER-relevant activities, which are the main driver of advances in the field. Discussion is made of the importance of efforts in these and other areas, considering some of the more pressing needs and requirements. In some cases, they call for a more efficient and coordinated use of the scarce resources available.

Computational Molecular Engineering as an Emerging Technology in Process Engineering

Abstract The present level of development of molecular force field methods is assessed from the point of view of simulation-based engineering, outlining the immediate perspective for further development and highlighting the newly emerging discipline of “Computational Molecular Engineering (CME)” which makes basic research in soft matter physics fruitful for industrial applications. Within the coming decade, major breakthroughs can be reached if a research focus is placed on processes at interfaces, combining aspects where an increase in the accessible length and time scales due to massively parallel high-performance computing will lead to particularly significant improvements.

Construction and optimization techniques for high order schemes for the two-dimensional wave equation

With the advent of high performance parallel computing, audio rate room auralization using finite difference time domain methods is becoming possible in a reasonable computation time. Yet, there are still deficiencies in the methods which are used for this purpose, particularly with regard to minimizing numerical dispersion over the full range of audible frequencies. This paper is concerned with construction techniques for families of methods for the test case of the 2D wave equation. Such methods are explicit, can be of very high accuracy, and operate over a small local stencil. Such schemes can be attractive in a parallel computation environment. As such methods will depend, invariably, on a set of free parameters, including the Courant number, a major concern is optimization. The remainder of this paper approaches the problem of setting up such an optimization problem in terms of various constraints and a suitable cost function. Some of the constraints follow from consistency, stability, isotropy and accuracy of the resulting scheme, and others from perceptual considerations peculiar to audio. Simulation results will be presented.

Bayesian updating via bootstrap filtering combined with data-driven polynomial chaos expansions: methodology and application to history matching for carbon dioxide storage in geological formations

Model calibration and history matching are important techniques to adapt simulation tools to real-world systems. When prediction uncertainty needs to be quantified, one has to use the respective statistical counterparts, e.g., Bayesian updating of model parameters and data assimilation. For complex and large-scale systems, however, even single forward deterministic simulations may require parallel high-performance computing. This often makes accurate brute-force and nonlinear statistical approaches infeasible. We propose an advanced framework for parameter inference or history matching based on the arbitrary polynomial chaos expansion (aPC) and strict Bayesian principles. Our framework consists of two main steps. In step 1, the original model is projected onto a mathematically optimal response surface via the aPC technique. The resulting response surface can be viewed as a reduced (surrogate) model. It captures the model’s dependence on all parameters relevant for history matching at high-order accuracy. Step 2 consists of matching the reduced model from step 1 to observation data via bootstrap filtering. Bootstrap filtering is a fully nonlinear and Bayesian statistical approach to the inverse problem in history matching. It allows to quantify post-calibration parameter and prediction uncertainty and is more accurate than ensemble Kalman filtering or linearized methods. Through this combination, we obtain a statistical method for history matching that is accurate, yet has a computational speed that is more than sufficient to be developed towards real-time application. We motivate and demonstrate our method on the problem of CO2 storage in geological formations, using a low-parametric homogeneous 3D benchmark problem. In a synthetic case study, we update the parameters of a CO2/brine multiphase model on monitored pressure data during CO2 injection.

Epidemic simulation of a large-scale social contact network on GPU clusters

Computer simulation is an important tool for studying epidemic dynamics. Owing to the scales and runtime speed requirements, the simulations of large-scale pandemic diseases such as severe acute respiratory syndrome (SARS) and H1N1 influenza usually require high-performance parallel simulation or computation. Previous works on the parallel simulations of large-scale epidemic were implemented on traditional general purpose CPU-based platforms or clusters. As more and more high-performance computation clusters are being built with so-called general-purpose graphics cards, this paper presents the implementation and optimization techniques for social contact network-based large-scale epidemic simulation on GPU clusters. Compared with previous works, this paper focuses on (1) how to efficiently implement the contact network-based parallel epidemic simulation on GPUs, and (2) how to hide communication latencies between processing nodes to improve scalability. Our proposed techniques are implemented and tested on a commodity cluster whose processing nodes are equipped with GPGPUs. The experimental results show that, for the simulation of 20 million individuals and 1.2 billion host contacts on 80 nodes, the execution on GPUs can achieve 7.4×−11.7× speedup over the execution on CPUs.

Particle-In-Cell Monte Carlo Collision Model on GPU—Application to a Low-Temperature Magnetized Plasma

The particle-in-cell (PIC) algorithm for the simulation of charged-particle kinetics in plasmas is a very resource consuming method, and high-performance parallel computing is required for practical problems. Graphics processing units (GPUs) are powerful low-cost parallel systems that can be used for intensive computations. We have developed a PIC Monte Carlo collision (MCC) model of a low temperature magnetized plasma using GPUs. We describe how each part of the PIC MCC model is implemented on the GPU and show how particles are dynamically managed. The computational cost of the PIC MCC model on the GPU is compared with a standard PIC MCC model running on a single central processing unit (CPU). We show that speedup can reach from 10 to 20 times compared with a sequential code running on a CPU, depending on the number of cells and particles considered. The results are illustrated with the example of plasma transport across a magnetic filter similar to that of a negative-ion source for the neutral beam injector of fusion devices.

Sequestration of CO<sub>2</sub> in Geological Media: A Case Study of the Saline Reservoir Using Parallel Computing for Multi-Well Injection

Carbon-Dioxide Capture and Storage (CCS) in deep saline aquifers is one of the most feasible techniques for reducing anthropogenic emission of carbon dioxide (CO2). In this paper, a high-performance parallel computing is used to simulate the large-scale and long-term CO2 geologic storage in the saline aquifer (Sleipner Vest field in the Norwegian) based on the ECO2N module of the flow/transport simulator TOUGH2-MP, which is the parallel version of TOUGH2 implemented by the MPI. We have developed a complex three-dimensional heterogeneous model to study the spatial and temporal distribution and storage of CO2 injection into the sands of the Utsira formation, at the Sleipner Vest field in the Norway. Simulation results demonstrate that the high-speed parallel computing enhanced the capability on handling the large-scale model and the long-term studies. Furthermore, in order to avoid the problems of overpressure in the saline reservoir, the case study employs multi-well (ten-well) injection model, which has been proven to be able to reduce the reservoir pressure effectively when compared to the single-well injection model.

J011011 高性能並列流体計算のための大規模格子生成手法の開発([J011-01]次世代スパコン「京」が拓くシミュレーションの新時代(1))

J011011 高性能並列流体計算のための大規模格子生成手法の開発([J011-01]次世代スパコン「京」が拓くシミュレーションの新時代(1))

Seismic Source Characteristics of Nuclear and Chemical Explosions in Granite from Hydrodynamic Simulations

Seismic source characteristics of low-yield (0.5–5 kt) underground explosions are inferred from hydrodynamic simulations using a granite material model on high-performance (parallel) computers. We use a non-linear rheological model for granite calibrated to historical near-field nuclear test data. Equivalent elastic P-wave source spectra are derived from the simulated hydrodynamic response using reduced velocity potentials. Source spectra and parameters are compared with the models of Mueller and Murphy (Bull Seism Soc Am 61:1675–1692, 1971, hereafter MM71) and Denny and Johnson (Explosion source phenomenology, pp 1–24, 1991, hereafter DJ91). The source spectra inferred from the simulations of different yields at normal scaled depth-of-burial (SDOB) match the MM71 spectra reasonably well. For normally buried nuclear explosions, seismic moments are larger for the hydrodynamic simulations than MM71 (by 25 %) and for DJ91 (by over a factor of 2), however, the scaling of moment with yield across this low-yield range is consistent for our calculations and the two models. Spectra from our simulations show higher corner frequencies at the lower end of the 0.5–5.0 kt yield range and stronger variation with yield than the MM71 and DJ91 models predict. The spectra from our simulations have additional energy above the corner frequency, probably related to non-linear near-source effects, but at high frequencies the spectral slopes agree with the f −2 predictions of MM71. Simulations of nuclear explosions for a range of SDOB from 0.5 to 3.9 show stronger variations in the seismic moment than predicted by the MM71 and DJ91 models. Chemical explosions are found to generate higher moments by a factor of about two compared to nuclear explosions of the same yield in granite and at normal depth-of-burial, broadly consistent with comparisons of nuclear and chemical shots at the US Nevada Test Site (Denny, Proceeding of symposium on the non-proliferation experiment, Rockville, Maryland, 1994). For all buried explosions, the region of permanent deformation and material damage is not spherical but extends along the free surface above and away from the source. The effect of damage induced by a normally buried nuclear explosion on seismic radiation is explored by comparing the motions from hydrodynamic simulations with those for point-source elastic Green’s functions. Results show that radiation emerging at downward takeoff angles appears to be dominated by the expected isotropic source contribution, while at shallower angles the motions are complicated by near-surface damage and cannot be represented with the addition of a simple secondary compensated linear vector dipole point source above the shot point. The agreement and differences of simulated source spectra with the MM71 and DJ91 models motivates the use of numerical simulations to understand observed motions and investigate seismic source features for underground explosions in various emplacement media and conditions, including non-linear rheological effects such as material strength and porosity.

Using high performance techniques to accelerate demand-driven hydraulic solvers

Computer models of water distribution networks are commonly used to simulate large systems under complex dynamic scenarios. These models normally use so-called demand-driven solvers, which determine the nodal pressures and pipe flow rates that correspond to specified nodal demands. This paper investigates the use of data parallel high performance computing (HPC) techniques to accelerate demand-driven hydraulic solvers. The sequential code of the solver implemented in the CWSNet library is analysed to understand which computational blocks contribute the most to the total computation time of a hydraulic simulation. The results obtained show that, contrary to popular belief, the linear solver is not the code block with the highest impact on the simulation time, but the pipe head loss computation. Two data parallel HPC techniques, single instruction multiple data (SIMD) operations and general purpose computation on graphics processing units (GPGPU), are used to accelerate the pipe head loss computation and linear algebra operations in new implementations of the hydraulic solver of CWSNet library. The results obtained on different network models show that the use of this techniques can improve significantly the performance of a demand-driven hydraulic solver.

A Provenance-based Adaptive Scheduling Heuristic for Parallel Scientific Workflows in Clouds

In the last years, scientific workflows have emerged as a fundamental abstraction for structuring and executing scientific experiments in computational environments. Scientific workflows are becoming increasingly complex and more demanding in terms of computational resources, thus requiring the usage of parallel techniques and high performance computing (HPC) environments. Meanwhile, clouds have emerged as a new paradigm where resources are virtualized and provided on demand. By using clouds, scientists have expanded beyond single parallel computers to hundreds or even thousands of virtual machines. Although the initial focus of clouds was to provide high throughput computing, clouds are already being used to provide an HPC environment where elastic resources can be instantiated on demand during the course of a scientific workflow. However, this model also raises many open, yet important, challenges such as scheduling workflow activities. Scheduling parallel scientific workflows in the cloud is a very complex task since we have to take into account many different criteria and to explore the elasticity characteristic for optimizing workflow execution. In this paper, we introduce an adaptive scheduling heuristic for parallel execution of scientific workflows in the cloud that is based on three criteria: total execution time (makespan), reliability and financial cost. Besides scheduling workflow activities based on a 3-objective cost model, this approach also scales resources up and down according to the restrictions imposed by scientists before workflow execution. This tuning is based on provenance data captured and queried at runtime. We conducted a thorough validation of our approach using a real bioinformatics workflow. The experiments were performed in SciCumulus, a cloud workflow engine for managing scientific workflow execution.

A hybrid Laplace transform finite analytic method for solving transport problems with large Peclet and Courant numbers

In this study, the authors develop a hybrid Laplace transform finite analytic method (LTFAM) to solve the advection–dispersion equations with large Peclet and Courant numbers. The finite analytic method with a hybrid Laplace transform can incorporate the temporal variable into the numerical scheme and effectively control the numerical dispersion and oscillation at solute sharp fronts. Since the conventional numerical methods use a large amount of time steps to iterate to the specified time, they may lead to an accumulation of computation errors from each iteration step. Instead of using many fine time steps to satisfy the condition of Courant numbers less than 1 for the conventional numerical methods, the LTFAM algorithm uses a one-step approach to compute the solute concentrations at any specified time with stable numerical solutions. The derived LTFAM algorithm is verified with two numerical simulation examples against the analytical solutions. The numerical results of the LTFAM match the analytical solutions very well, especially for solute transport in the advection-dominated cases. The developed algorithm in this paper can save a large amount of simulating time and improve the computational accuracy. Furthermore, because the solutions of the LTFAM for a set of specified times can be obtained separately in the Laplace space, independence of each time step implies that the LTFAM is well-suited for executing on high performance parallel computers. This algorithm facilitates the long-term predictions of contaminant transport in the kilometer-scale field sites.

Evaluating High-Performance Computing on Google App Engine

An experimental approach employs the Google App Engine (GAE) for high-performance parallel computing. A generic master-slave framework enables fast prototyping and integration of parallel algorithms that are transparently scheduled and executed on the Google cloud infrastructure. Compared to Amazon Elastic Compute Cloud (EC2), GAE offers lower resource-provisioning overhead and is cheaper for jobs shorter than one hour. Experiments demonstrated good scalability of a Monte Carlo simulation algorithm. Although this approach produced important speedup, two main obstacles limited its performance: middleware overhead and resource quotas.

The IBM Blue Gene/Q Compute Chip

Blue Gene/Q aims to build a massively parallel high-performance computing system out of power-efficient processor chips, resulting in power-efficient, cost-efficient, and floor-space- efficient systems. Focusing on reliability during design helps with scaling to large systems and lowers the total cost of ownership. This article examines the architecture and design of the Compute chip, which combines processors, memory, and communication functions on a single chip.

Tuple switching network—When slower may be better

This paper reports an application dependent network design for extreme scale high performance computing (HPC) applications. Traditional scalable network designs focus on fast point-to-point transmission of generic data packets. The proposed network focuses on the sustainability of high performance computing applications by statistical multiplexing of semantic data objects. For HPC applications using data-driven parallel processing, a tuple is a semantic object. We report the design and implementation of a tuple switching network for data parallel HPC applications in order to gain performance and reliability at the same time when adding computing and communication resources. We describe a sustainability model and a simple computational experiment to demonstrate extreme scale application’s sustainability with decreasing system mean time between failures (MTBF). Assuming three times slowdown of statistical multiplexing and 35% time loss per checkpoint, a two-tier tuple switching framework would produce sustained performance and energy savings for extreme scale HPC application using more than 1024 processors or less than 6 hour MTBF. Higher processor counts or higher checkpoint overheads accelerate the benefits.

A new parallel association rule mining algorithm on distributed shared memory system

Frequent itemset finding is the most time consuming step in analysing large transactional databases. The use of sequential algorithms cannot give analytical ability for such very large databases especially in terms of run-time performance. Therefore, we must rely on high performance parallel computing. In this paper, we present a new parallel algorithm for frequent itemset mining, called 'HorVertical' algorithm. This algorithm introduces a new database partitioning called 'HorVertical' partitioning. This technique in partitioning the database reduces the dependency in the parallel computation and gives new properties to reduce the computations. The algorithm passes the database only one time and starts a new stage with the finished itemsets while some other itemsets in the same stage have not been finished yet. We present the result on the performance of our algorithm on various databases, and compare it against well known algorithms.

P3DFFT: A Framework for Parallel Computations of Fourier Transforms in Three Dimensions

Fourier and related transforms is a family of algorithms widely employed in diverse areas of computational science, notoriously difficult to scale on high-performance parallel computers with large number of processing elements (cores). This paper introduces a popular software package called P3DFFT implementing Fast Fourier Transforms (FFT) in three dimensions (3D) in a highly efficient and scalable way. It overcomes a well-known scalability bottleneck of 3D FFT implementations by using two-dimensional domain decomposition. Designed for portable performance, P3DFFT achieves excellent timings for a number of systems and problem sizes. On Cray XT5 system P3DFFT attains 45% efficiency in weak scaling from 128 to 65,536 computational cores. Library features include Fourier and Chebyshev transforms, Fortran and C interfaces, in- and out-of-place transforms, uneven data grids, single and double precision. P3DFFT is available as open source at http://code.google.com/p/p3dfft/. This paper discusses P3DFFT implementation and performance in a way that helps guide the user in making optimal choices for parameters of their runs.

FACC: A Novel Finite Automaton Based on Cloud Computing for the Multiple Longest Common Subsequences Search

Searching for the multiple longest common subsequences (MLCS) has significant applications in the areas of bioinformatics, information processing, and data mining, and so forth, Although a few parallel MLCS algorithms have been proposed, the efficiency and effectiveness of the algorithms are not satisfactory with the increasing complexity and size of biologic data. To overcome the shortcomings of the existing MLCS algorithms, and considering that MapReduce parallel framework of cloud computing being a promising technology for cost‐effective high performance parallel computing, a novel finite automaton (FA) based on cloud computing called FACC is proposed under MapReduce parallel framework, so as to exploit a more efficient and effective general parallel MLCS algorithm. FACC adopts the ideas of matched pairs and finite automaton by preprocessing sequences, constructing successor tables, and common subsequences finite automaton to search for MLCS. Simulation experiments on a set of benchmarks from both real DNA and amino acid sequences have been conducted and the results show that the proposed FACC algorithm outperforms the current leading parallel MLCS algorithm FAST‐MLCS.

A new electromagnetic particle-in-cell model with adaptive mesh refinement for high-performance parallel computation

A new electromagnetic particle-in-cell (EMPIC) model with adaptive mesh refinement (AMR) has been developed to achieve high-performance parallel computation in distributed memory system. For minimizing the amount and frequency of inter-processor communications, the present study uses the staggering grid scheme with the charge conservation method, which consists only of the local operations. However, the scheme provides no numerical damping for electromagnetic waves regardless of the wavenumber, which results in significant noise in the refinement region that eventually covers over physical signals. In order to suppress the electromagnetic noise, the present study introduces a smoothing method which gives numerical damping preferentially for short wavelength modes. The test simulations show that only a weak smoothing results in drastic reduction in the noise, so that the implementation of the AMR is possible in the staggering grid scheme. The computational load balance among the processors is maintained by a new method termed the adaptive block technique for the domain decomposition parallelization. The adaptive block technique controls the subdomain (block) structure dynamically associated with the system evolution, such that all the blocks have almost the same number of particles. The performance of the present code is evaluated for the simulations of the current sheet evolution. The test simulations demonstrate that the usage of the adaptive block technique as well as the staggering grid scheme enhances significantly the parallel efficiency of the AMR-EMPIC model.

Toward real-time Monte Carlo simulation using a commercial cloud computing infrastructure**This work was presented in part at the 2010 Annual Meeting of the American Association of Physicists in Medicine (AAPM), Philadelphia, PA.

Monte Carlo (MC) methods are the gold standard for modeling photon and electron transport in a heterogeneous medium; however, their computational cost prohibits their routine use in the clinic. Cloud computing, wherein computing resources are allocated on-demand from a third party, is a new approach for high performance computing and is implemented to perform ultra-fast MC calculation in radiation therapy. We deployed the EGS5 MC package in a commercial cloud environment. Launched from a single local computer with Internet access, a Python script allocates a remote virtual cluster. A handshaking protocol designates master and worker nodes. The EGS5 binaries and the simulation data are initially loaded onto the master node. The simulation is then distributed among independent worker nodes via the message passing interface, and the results aggregated on the local computer for display and data analysis. The described approach is evaluated for pencil beams and broad beams of high-energy electrons and photons. The output of cloud-based MC simulation is identical to that produced by single-threaded implementation. For 1 million electrons, a simulation that takes 2.58 h on a local computer can be executed in 3.3 min on the cloud with 100 nodes, a 47× speed-up. Simulation time scales inversely with the number of parallel nodes. The parallelization overhead is also negligible for large simulations. Cloud computing represents one of the most important recent advances in supercomputing technology and provides a promising platform for substantially improved MC simulation. In addition to the significant speed up, cloud computing builds a layer of abstraction for high performance parallel computing, which may change the way dose calculations are performed and radiation treatment plans are completed.