Large-scale Parallel Computers Research Articles

TT-HEALpix: A New Data Indexing Strategy for Efficient Cross-match of Large-scale Astronomical Catalogs

Cross-matching is an indispensable operation in the data preparation, analysis, and research processes of multi-band astronomy and time-domain astronomy. Multi-catalog time-series data reconstruction is an important part of time-domain astronomy. In the large-scale distributed reconstruction process, boundary problems have always affected the accuracy of time-series data. To optimize these boundary problems and improve data precision, this paper proposes a new hybrid astronomical data indexing method called Translated Transformation based HEALPix Dual Index (TT-HEALPix). Under the reasonable Healpix division level, by translation transformation, the two indexes before and after the transformation form a unique pseudo-hybrid index strategy, which not only retains the advantages of the hybrid index scheme suitable for large-scale parallel computing, but also compensates for its shortage of high omission at the block boundary position. Based on TT-HEALPix, this paper completes the multi-catalog time-series reconstruction process on the Spark platform and compares it with the HEALPix+HTM hybrid indexing strategy. The experiments demonstrate that TT-HEALPix has significant advantages over the traditional HEALPix+HTM hybrid indexing method in terms of data accuracy and cross-matching efficiency. At level 9 of the Healpix index, TT-HEALPix achieves a 6%–19% improvement in cross-matching efficiency in a distributed environment compared to HEALPix+HTM. In terms of data accuracy, for the AST3-II dataset at level 9, TT-HEALPix has 62.2% accuracy improvement over HEALPix and 45.5% improvement over HEALPix+HTM. In conclusion, the proposed novel indexing strategy, TT-HEALPix, is better suited to the efficiency and accuracy requirements of cross-match.

Hydrological Simulation Study in Gansu Province of China Based on Flash Flood Analysis

The calibration and validation of hydrological model simulation performance and model applicability evaluation in Gansu Province is the foundation of the application of the flash flood early warning and forecasting platform in Gansu Province. It is difficult to perform the simulation for Gansu Province due to the fact that it covers a wide range, from north to south, with multiple climate types and diverse landforms. The China Flash Flood Hydrological Model (CNFF) was implemented in this study. A total of 11 model clusters and 289 distributed hydrological models were divided based on hydrology, climate, and land-use factors, among others. A spatiotemporally mixed runoff method and the Event-Specific Geomorphological Instantaneous Unit Hydrograph (GIUH) were applied based on large-scale fast parallel computation. To improve model calibration and validation efficiency, the RSA method (Regionalized Sensitivity Analysis) was used for CNFF model parameter sensitivity analysis, which could reduce the number of model parameters that need to be adjusted during the calibration period. Based on the model sensitivity analysis results, the CNFF was established in Gansu Province to simulate flood events in eight representative watersheds. The average NSE, REQ, and ET were 0.76 and 0.73, 9.1% and 12.6%, and 1.2 h and 1.7 h, respectively, in the calibration and validation period. In general, the CNFF model shows a good performance in multiple temporal and spatial scales, thus providing a scientific basis for flash flood early warning and analysis in Gansu Province.

Low consumption automatic discovery protocol for DDS-based large-scale distributed parallel computing

DDS (Data Distribution Service) is an efficient communication specification for distributed parallel computing. However, as the scale of computation expands, high network load and memory consumption consistently limit its performance. This paper proposes a low consumption automatic discovery protocol to improve DDS in large-scale distributed parallel computing. Firstly, an improved Bloom Filter called TBF (Threshold Bloom Filter) is presented to compress the data topic. Then it is combined with the SDP(Simple Discovery Protocol) to reduce the consumption of the automatic discovery process in DDS. On this basis, data publication and subscription between the distributed computing nodes are matched using binarization threshold θ and decision threshold T , which can be obtained through iterative optimization algorithms. Experiment results show that the SDPTBF can guarantee higher transmission accuracy while reducing network load and memory consumption, and therefore improve the performance of DDS-based large-scale distributed parallel computing.

Stability analysis of a deep and large open pit based on fine geological modeling and large-scale parallel computing: a case study of Fushun West Open-pit Mine

Accurate input of geological elements is essential for evaluating or predicting natural hazards such as subsidence, landslides, and earthquakes. This paper proposes an approach to carry out an open pit’s overall and whole-process mechanical analysis with complex geological conditions, using precise modeling and large-scale parallel calculation techniques. Taking the Fushun West Open-pit Mine (the largest open-pit coal mine in Asia) as an example, through the elaborate multi-method geological investigation, the interfaces of interbedded shales and mudstones, the unloading zones, and the small structures were identified, a detailed 3D geological model was built and finely meshed in full-size with 100 million degrees of freedom, large-scale parallel numerical simulation was then performed, the results agree well with the InSAR monitoring data and in situ observations. Besides, the simulation can replicate the landslides in recent years. Through the simulation, it is possible to locate the potential landslide area, and targeted backfilling schemes for stability treatment were put forward and further simulated. The results indicate that the proposed approach can more effectively and reliably evaluate the Fushun West Open-pit Mine’s overall slope stability and closure plan.

Scalability of Viscoelastic Fluid Solvers Based on OpenFOAM-PETSc Framework in Large-Scale Parallel Computing

Enormous advances in physics of complex fluids/soft matter over last decades have rapidly transformed traditional industrial sectors in foods, personal care products, pharmaceuticals, paints, lubricants, ceramics, polymers, liquid crystals, high performance fibers, oil exploration and production into a digital era of formulation design and precision control over processing conditions from molecular viewpoint, and fertilizing a new industrial revolution. Development of high performance viscoelastic fluid solvers is of great significance for large scale digital manufacturing. In the present work, a portable and extensible scientific computing (PETSc) toolbox has been successfully integrated into the popular OpenFOAM CFD toolbox for carrying out large scale parallel computing of Turbulent Drag Reduction (TDR) and Elastic Turbulence (ET) in the isotropic turbulence flow. Its scalability has been evaluated and compared with the scalability of the OpenFOAM based viscoelastic fluid solvers. The results show that there are significant improvements.

Turbo-HB: A Sub-Millisecond Hybrid Beamforming Design for 5G mmWave Systems

Hybrid beamforming (HB) architecture has been widely considered for 5G mmWave systems. It reduces hardware complexity by allowing the number of RF chains to be far fewer than the number of antennas. A major practical challenge for HB is to obtain a beamforming solution in real-time. In 5G NR, new frame structures with short TTIs are employed to support mmWave communications. Under such frame structures, it is necessary to obtain a beamforming solution with a time resolution varying from 1 ms to 125 us -- an extremely stringent time requirement considering the complexity involved in HB. In this paper, we present the design and implementation of Turbo-HB -- a novel beamforming design under the HB architecture that is capable of offering the beamforming matrices in less than 500 us. The key ideas of Turbo-HB include: (i) reducing the complexity of computation-intensive SVD operations by exploiting channel sparsity at mmWave frequencies, and (ii) achieving large-scale parallel computation with minimal memory access. We implement Turbo-HB on an off-the-shelf Nvidia GPU and conduct extensive experiments. Our experimental results demonstrate that Turbo-HB can obtain a beamforming solution in 500 us for up to 100 RBs and 10 MU-MIMO users on each RB while offering competitive throughput performance compared to state-of-the-art (non-real-time) algorithms.

A robust parallel algorithm for Eulerian-Lagrangian method crossing sliding non-conformal interfaces

Numerical simulations of flows with dilute disperse phase play an important role in a wide range of industrial applications. The most commonly used technique for solving such problems is the Eulerian-Lagrangian approach, in which individual particles are tracked inside the domain. However, the use of sliding and moving mesh in complex engineering applications poses more challenges for the parallelization of the Eulerian-Lagrangian method in the two-way coupling. In this context, this manuscript presents the parallel algorithm for tracking Lagrangian particles crossing sliding non-conformal interfaces in a mixed Eulerian-Lagrangian CFD algorithm, which is based on a parallel supermesh method introduced recently for the handling of sliding mesh in large-scale parallel computing (Y. Xiao and P.J. Ming, Jour. Computat. Phys., 2022). In the proposed parallel algorithm, critical problems such as geometric calculation for particles across non-conformal mesh interfaces and the dynamic communication for particle data can be handled based on the one-to-one addressing constructed by the supermesh method introduced in the author's previous work. Further, a robust and easy-to-implement algorithm is also provided for particle communication caused by mesh moving. All methodologies are integrated and implemented for parallel computation obtaining good robustness and consistency. Specific examples are solved to examine the performance of this parallel strategy when applied to practical engineering problems.

The Peridigm Meshfree Peridynamics Code

Peridigm is a meshfree peridynamics code written in C++ for use on large-scale parallel computers. It was originally developed at Sandia National Laboratories and is currently managed as an open-source, community driven software project. Its primary features include bond-based, state-based, and non-ordinary state-based constitutive models, bond failure laws, contact, and support for explicit and implicit time integration. To date, Peridigm has been used primarily by methods developers focused on solid mechanics and material failure. Peridigm utilizes foundational software components from Sandia’s Trilinos project and was designed for extensibility. This paper provides an overview of the solution methods implemented in Peridigm, a discussion of its software infrastructure, and demonstrates the use of Peridigm for the solution of several example problems.

Topology optimization using the lattice Boltzmann method for unsteady natural convection problems

This paper proposes a density-based topology optimization method for natural convection problems using the lattice Boltzmann method (LBM). As the LBM can be developed as a completely explicit scheme, its attractive features over the traditional ones, such as the finite element method, are (1) suitability for solving unsteady flow problems and (2) scalability for large-scale parallel computing. We develop an LBM code for solving unsteady natural convection problems and provide its sensitivity analysis based on the so-called adjoint lattice Boltzmann method. Notably, the adjoint equation is derived from the discrete particle velocity Boltzmann equation and can be solved similarly to the original LBM concerning unsteady natural convection problems. We first show that the proposed method can produce similar results to the previous work in a steady-state natural convection problem. We then demonstrate the efficacy of the proposed method through 2D numerical examples concerning unsteady natural convection. As a large-scale problem, we tackle a 3D unsteady natural convection problem on a parallel supercomputer. All the developed codes written in C++ are available at https://github.com/PANFACTORY/PANSLBM2.git.

A Scalable Algorithm for Shape Optimization with Geometric Constraints in Banach Spaces

This work develops an algorithm for PDE-constrained shape optimization based on Lipschitz transformations. Building on previous work in this field, the $p$-Laplace operator is utilized to approximate a descent method for Lipschitz shapes. In particular, it is shown how geometric constraints are algorithmically incorporated avoiding penalty terms by assigning them to the subproblem of finding a suitable descent direction. A special focus is placed on the scalability of the proposed methods for large scale parallel computers via the application of multigrid solvers. The preservation of mesh quality under large deformations, where shape singularities have to be smoothed or generated within the optimization process, is also discussed. It is shown that the interaction of hierarchically refined grids and shape optimization can be realized by the choice of appropriate descent directions. The performance of the proposed methods is demonstrated for energy dissipation minimization in fluid dynamics applications.

A globally convergent smoothing Newton method for the second order cone complementarity approach of elastoplasticity problems

A new algorithm for solving small-deformation elastoplasticity problems as second order cone complementarity problems (SOCCPs) is presented. The SOCCP is derived from the Hellinger–Reissner variational principle, and then solved with a smoothing Newton method that is modified from the globally convergent Qi-Sun-Zhou method. The method is compared with two well-established methods: the return-mapping method and the conic programming (CP) approach with primal–dual interior-point method. Numerical results indicate that the proposed method shares common characteristics of CP class methods and offers better convergence efficiency in some cases. Moreover, it is shown that the proposed method can be imbedded in the framework of conventional finite element codes with only a few modifications, which makes it easy-to-implement and adaptive to large-scale parallel computing.

A concurrent multiscale method based on smoothed molecular dynamics for large-scale parallel computation at finite temperature

A concurrent atomic-to-continuum multiscale method for finite-temperature simulation with large-scale parallel computation is developed. Seamless and stable coupling between molecular dynamics (MD) method and continuum-based material point method (MPM) is achieved with the aid of mesoscopic smoothed molecular dynamics (SMD) method. Novel techniques are proposed to filter spurious high-frequency reflections and bridge different thermal descriptions in atomistic and continuum models by adopting the efficient Markovian generalized Langevin equation (GLE) with newly designed drift matrix. Besides, an effective relaxation strategy is developed to fully relax the multiscale model to thermodynamic equilibrium state. The concurrent multiscale method can be effectively parallelized within the parallel computation framework of particle-grid dual discretization, and its efficiency when executing on large-scale computation clusters is analyzed in detail. The contact-sliding and particle impact examples are calculated with the proposed method to demonstrate its superiority as an effective numerical tool in the practical applications.

Developing an early-warning system through robotic process automation: Are intelligent tutoring robots as effective as human teachers?

ABSTRACT Artificial intelligence aims to restructure and process re-engineering education and teaching processes and accelerate the evolution of the whole education system from information to intelligence. Robotic Process Automation (RPA) robots learn by observing people at work, analyzing user processes repeatedly, and adjusting or correcting automated processes. By using Natural Language Processing (NLP) and Machine Learning (ML), knowledge representation, inference, large-scale parallel computing, and Rapid Domain Adaptation, RPA robots can automatically extract the data needed for decision-making and continuously learn from users’ feedback. We have used RPA and predictive analytics to provide distance learning students with the Intelligent Tutoring Robot (ITR), which can provide an automatic response. By optimizing the ITR in the above context, we have examined the feasibility of transforming a prediction model, using a student learning database, into an early-warning system. This article adopts the randomized control-group pretest-posttest design, dividing 123 students into a control group to describe interactions between ITR and students and experimental groups to describe interactions between human teachers and students. The findings present no significant difference between the control and the experimental groups in terms of academic performance, however higher average marks were achieved in the former group.

Parallel Simulation of Crowd Multi-Cell Occupancy and Velocity Variety

A parallel algorithm is proposed for modeling of crowd multi-cell occupancy and velocity variety. Methods are devised to solve issues such as space competitions so that each pedestrian can autonomously determines his or her own movement. As pedestrians are modeled independently and simultaneously, the proposed algorithm can be easily implemented into heterogeneous computing frameworks such as OpenCL, CUDA etc, which support large scale parallel computation, and preserve these frameworks’ high performance while the issues are efficiently solved. In the section of numerical experiments, the proposed algorithm is firstly validated through two different experiments. Then a quite thorough investigation of impacts of multi-cell occupancy and velocity variety upon crowd macroscopic dynamics is applied.

A high fidelity general purpose 3-D Monte Carlo particle transport program JMCT3.0

JMCT is a large-scale, high-fidelity, three-dimensional general neutron–photon–electron–proton transport Monte Carlo software system. It was developed based on the combinatorial geometry parallel infrastructure JCOGIN and the adaptive structured mesh infrastructure JASMIN. JMCT is equipped with CAD modeling and visualizes the image output. It supports the geometry of the body and the structured/unstructured mesh. JMCT has most functions, variance reduction techniques, and tallies of the traditional Monte Carlo particle transport codes. Two energy models, multi-group and continuous, are provided. In recent years, some new functions and algorithms have been developed, such as Doppler broadening on-the-fly (OTF), uniform tally density (UTD), consistent adjoint driven importance sampling (CADIS), fast criticality search of boron concentration (FCSBC) domain decomposition (DD), adaptive control rod moving (ACRM), and random geometry (RG) etc. The JMCT is also coupled with the discrete ordinate SN code JSNT to generate source-biasing factors and weight-window parameters. At present, the number of geometric bodies, materials, tallies, depletion zones, and parallel processors are sufficiently large to simulate extremely complicated device problems. JMCT can be used to simulate reactor physics, criticality safety analysis, radiation shielding, detector response, nuclear well logging, and dosimetry calculations etc. In particular, JMCT can be coupled with depletion and thermal-hydraulics for the simulation of reactor nuclear-hot feedback effects. This paper describes the progress in advanced modeling, high-performance numerical simulation of particle transport, multiphysics coupled calculations, and large-scale parallel computing.

A load-decoupling parallel strategy based on shared memory architecture for DSMC to simulate near-continuum gases

Direct Simulation Monte Carlo (DSMC) method is gradually being applied to simulate near-continuum flowfields. However, the large amount of computation hinders its application. Parallelization is the primary way to solve this problem. Nevertheless, when the Knudsen number (Kn) is less than 1.0×10−2, it is hard to gain good load balance while parallelized since the overall computational load of DSMC is challenging to determine and also changes with cases. This paper proposed a fine-grain parallel strategy based on the shared memory architecture. This strategy is designed to gain good load balance through reasonably decoupling computational load, which means that the load of each module of the program is treated separately, not as a whole. Implemented through Open Multi-Processing (OpenMP), a “dual-decomposition” method is adopted to achieve the load-decoupling strategy. In the “dual-decomposition” method, the computation domain is decomposed with a different weight according to whether one module is particle-dominated or collision-dominated. Besides, a multi-segment partition algorithm with low overhead is proposed. Two cases, including one unsteady shock-bubble interaction (SBI) case and the other steady supersonic jet case, are used to verify our strategy's correctness and test the performance under different conditions. The results show the strategy newly proposed can reduce more than 70% load imbalance relative to our previous PartPlusColl strategy when Kn<1.0×10−2. Simultaneously, the results prove that our load-decoupling strategy can improve parallel efficiency by about 11.17%−21.80%, compared with Sparta [1]. This strategy sets the fundamental for improving parallel efficiency of large-scale parallel computing of DSMC.

Accurate quantum dynamics of the simplest isomerization system involving double-H transfer

We perform accurate quantum dynamics calculations on the isomerization of vinylidene-acetylene. Large-scale parallel computations are accomplished by an efficient theoretical scheme developed by our group, in which the basis functions are customized for the double-H transfer process. The A′1 and B″1 vinylidene and delocalization states are obtained. The peaks recently observed in the cryo-SEVI spectra are analyzed, and very good agreement for the energy levels is achieved between theory and experiment. The discrepancies of energy levels between our calculations and recent experimental cryo-SEVI spectra are of similar magnitudes to the experimental error bars, or ≤30 cm−1 excluding those involving the excitation of the CCH2 scissor mode. A kind of special state, called the isomerization state, is revealed and reported, which is characterized by large probability densities in both vinylidene and acetylene regions. In addition, several states dominated by vinylidene character are reported for the first time. The present work would contribute to the understanding of the double-H transfer.

Fast Penalized Regression and Cross Validation for Tall Data with the oem Package

A large body of research has focused on theory and computation for variable selection techniques for high dimensional data. There has been substantially less work in the big "tall" data paradigm, where the number of variables may be large, but the number of observations is much larger. The orthogonalizing expectation maximization (OEM) algorithm is one approach for computation of penalized models which excels in the big tall data regime. The oem package is an efficient implementation of the OEM algorithm which provides a multitude of computation routines with a focus on big tall data, such as a function for out-of-memory computation, for large-scale parallel computation of penalized regression models. Furthermore, in this paper we propose a specialized implementation of the OEM algorithm for cross validation, dramatically reducing the computing time for cross validation over a naive implementation.

GPU-Accelerated Laplace Equation Model Development Based on CUDA Fortran

In this study, a CUDA Fortran-based GPU-accelerated Laplace equation model was developed and applied to several cases. The Laplace equation is one of the equations that can physically analyze the groundwater flows, and is an equation that can provide analytical solutions. Such a numerical model requires a large amount of data to physically regenerate the flow with high accuracy, and requires computational time. These numerical models require a large amount of data to physically reproduce the flow with high accuracy and require computational time. As a way to shorten the computation time by applying CUDA technology, large-scale parallel computations were performed on the GPU, and a program was written to reduce the number of data transfers between the CPU and GPU. A GPU consists of many ALUs specialized in graphic processing, and can perform more concurrent computations than a CPU using multiple ALUs. The computation results of the GPU-accelerated model were compared with the analytical solution of the Laplace equation to verify the accuracy. The computation results of the GPU-accelerated Laplace equation model were in good agreement with the analytical solution. As the number of grids increased, the computational time of the GPU-accelerated model gradually reduced compared to the computational time of the CPU-based Laplace equation model. As a result, the computational time of the GPU-accelerated Laplace equation model was reduced by up to about 50 times.

MPI detach — Towards automatic asynchronous local completion

When aiming for large-scale parallel computing, waiting time due to network latency, synchronization, and load imbalance are the primary opponents of high parallel efficiency. A common approach to hide latency with computation is the use of non-blocking communication. In the presence of a consistent load imbalance, synchronization cost is just the visible symptom of the load imbalance. Tasking approaches as in OpenMP, TBB, OmpSs, or C++20 coroutines promise to expose a higher degree of concurrency, which can be distributed on available execution units and significantly increase load balance. Available MPI non-blocking functionality does not integrate seamlessly into such tasking parallelization. In this work, we present a slim extension of the MPI interface to allow seamless integration of non-blocking communication with available concepts of asynchronous execution in OpenMP and C++. Using our concept allows to span task dependency graphs for asynchronous execution over the full distributed memory application. We furthermore investigate compile-time analysis necessary to transform an application using blocking MPI communication into an application integrating OpenMP tasks with our proposed MPI interface extension.