An MPI parallel DEM-IMB-LBM framework for simulating fluid-solid interaction problems

The high-resolution DEM-IMB-LBM model can accurately describe pore-scale fluid-solid interactions,but its potential for use in geotechnical engineering analysis has not been fully unleashed due to its prohibitive computational costs.To overcome this limitation,a message passing interface(MPI)parallel DEM-IMB-LBM framework is proposed aimed at enhancing computation efficiency.This framework utilises a static domain decomposition scheme,with the entire computation domain being decomposed into multiple subdomains according to predefined processors.A detailed parallel strategy is employed for both contact detection and hydrodynamic force calculation.In particular,a particle ID re-numbering scheme is proposed to handle particle transitions across sub-domain interfaces.Two benchmarks are conducted to validate the accuracy and overall performance of the proposed framework.Subsequently,the framework is applied to simulate scenarios involving multi-particle sedimentation and submarine landslides.The numerical examples effectively demonstrate the robustness and applicability of the MPI parallel DEM-IMB-LBM framework.

MPI环境下基于边界校正方法的海洋模型的并行化

基于区域分解技术,设计了一种针对海洋模型的边界校正方法,该方法不需要对原有海洋模型进行较大改动,只需要利用消息传递接口(message passing interface,MPI)编写子区域间的数据传输模块,即可实现原有海洋模型的并行化。相对于重新开发并行系统,该方法能够较快把原有模型的串行解决方案转化为并行解决方案,并且并行化后的海洋模型能够获得较高的加速比。

并行最短路径搜索算法的设计与实现

针对串行最短路径搜索算法本身固有的局限性,难以随着网络规模的增大而提高搜索速度的问题,设计并实现了一种基于并行Dijkstra思想的并行最短路径搜索算法,使算法复杂度由O(N2)减少到O(N2/p+N*(p-1)),提高了算法的效率。实验结果表明,该算法搜索速度快且性能稳定,当结点数目相当庞大时,算法的优越性更加明显。

Development of Ubiquitous Simulation Service Structure Based on High Performance Computing Technologies

The simulation field became essential in designing or developing new casting products and in improving manufacturing processes within limited time, because it can help us to simulate the nature of processing, so that developers can make ideal casting designs. To take the prior occupation at commercial simulation market, so many development groups in the world are doing their every effort. They already reported successful stories in manufacturing fields by developing and providing the high performance simulation technologies for multipurpose. But they all run at powerful desk-side computers by well-trained experts mainly, so that it is hard to diffuse the scientific designing concept to newcomers in casting field. To overcome upcoming problems in scientific casting designs, we utilized information technologies and full-matured hardware backbones to spread out the effective and scientific casting design mind, and they all were integrated into Simulation Portal on the web. It professes scientific casting design on the NET including ubiquitous access way represented by ＂Anyone, Anytime, Anywhere＂ concept for casting designs.

Parallel computation of unified finite-difference time-domain for underwater sound scattering

In this work, we treat scattering objects, water, surface and bottom in a truly unified manner in a parallel finitedifference time-domain （FDTD） scheme, which is suitable for distributed parallel computing in a message passing interface （MPI） programming environment. The algorithm is implemented on a cluster-based high performance computer system. Parallel computation is performed with different division methods in 2D and 3D situations. Based on analysis of main factors affecting the speedup rate and parallel efficiency, data communication is reduced by selecting a suitable scheme of task division. A desirable scheme is recommended, giving a higher speedup rate and better efficiency. The results indicate that the unified parallel FDTD algorithm provides a solution to the numerical computation of acoustic scattering.

Coupling analysis of transmission lines excited by space electromagnetic fields based on time domain hybrid method using parallel technique

We present a time domain hybrid method to realize the fast coupling analysis of transmission lines excited by space electromagnetic fields, in which parallel finite-difference time-domain (FDTD) method, interpolation scheme, and Agrawal model-based transmission line (TL) equations are organically integrated together. Specifically, the Agrawal model is employed to establish the TL equations to describe the coupling effects of space electromagnetic fields on transmission lines. Then, the excitation fields functioning as distribution sources in TL equations are calculated by the parallel FDTD method through using the message passing interface (MPI) library scheme and interpolation scheme. Finally, the TL equations are discretized by the central difference scheme of FDTD and assigned to multiple processors to obtain the transient responses on the terminal loads of these lines. The significant feature of the presented method is embodied in its parallel and synchronous calculations of the space electromagnetic fields and transient responses on the lines. Numerical simulations of ambient wave acting on multi-conductor transmission lines (MTLs), which are located on the PEC ground and in the shielded cavity respectively, are implemented to verify the accuracy and efficiency of the presented method.

Multi-Deme Parallel FGAs-Based Algorithm for Multitarget Tracking

For data association in multisensor and multitarget tracking, a novel parallel algorithm is developed to improve the efficiency and real-time performance of FGAs-based algorithm. One Cluster of Workstation （COW） with Message Passing Interface （MPI） is built. The proposed Multi-Deme Parallel FGA （MDPFGA） is run on the platform. A serial of special MDPFGAs are used to determine the static and the dynamic solutions of generalized m-best S-D assignment problem respectively, as well as target states estimation in track management. Such an assignment-based parallel algorithm is demonstrated on simulated passive sensor track formation and maintenance problem. While illustrating the feasibility of the proposed algorithm in multisensor multitarget tracking, simulation results indicate that the MDPFGAs-based algorithm has greater efficiency and speed than the FGAs-based algorithm.

Experiments in Parallelizing the Type IV Discrete Cosine Transform

The Fourier transform is very important to numerous applications in science and engineering. However, its usefulness is hampered by its computational expense. In this paper, in an attempt to develop a faster method for computing Fourier transforms, the authors present parallel implementations of two new algorithms developed for the type IV Discrete Cosine Transform （DCT-IV） which support the new interleaved fast Fourier transform method. The authors discuss the realizations of their implementations using two paradigms. The first involved commodity equipment and the Message-Passing Interface （MPI） library. The second utilized the RapidMind development platform and the Cell Broadband Engine （BE） processor. These experiments indicate that the authors＇ rotation-based algorithm is preferable to their lifting-based algorithm on the platforms tested, with increased efficiency demonstrated by their MPI implementation for large data sets. Finally, the authors outline future work by discussing an architecture-oriented method for computing DCT-IVs which promises further optimization. The results indicate a promising fresh direction in the search for efficient ways to compute Fourier transforms.

An efficient parallel algorithm for ocean circulation numerical model based on irregular rectangle decomposition scheme

A parallel algorithm of circulation numerical model based on message passing interface（MPI） is developed using serialization and an irregular rectangle decomposition scheme. Neighboring point exchange strategy（NPES） is adopted to further enhance the computational efficiency. Two experiments are conducted on HP C7000 Blade System, the numerical results show that the parallel version with NPES（PVN） produces higher efficiency than the original parallel version（PV）. The PVN achieves parallel efficiency in excess of 0.9 in the second experiment when the number of processors increases to 100, while the efficiency of PV decreases to 0.39 rapidly. The PVN of ocean circulation model is used in a fine-resolution regional simulation, which produces better results. The capability of universal implementation of this algorithm makes it applicable in many other ocean models potentially.

MPI-RCDD: A Framework for MPI Runtime Communication Deadlock Detection

The message passing interface (MPI) has become a de facto standard for programming models of highperformance computing, but its rich and flexible interface semantics makes the program easy to generate communication deadlock, which seriously affects the usability of the system. However, the existing detection tools for MPI communication deadlock are not scalable enough to adapt to the continuous expansion of system scale. In this context, we propose a framework for MPI runtime communication deadlock detection, namely MPI-RCDD, which contains three kinds of main mechanisms. Firstly, MPI-RCDD has a message logging protocol that is associated with deadlock detection to ensure that the communication messages required for deadlock analysis are not lost. Secondly, it uses the asynchronous processing thread provided by the MPI to implement the transfer of dependencies between processes, so that multiple processes can participate in deadlock detection simultaneously, thus alleviating the performance bottleneck problem of centralized analysis. In addition, it uses an AND⊕OR model based algorithm named AODA to perform deadlock analysis work. The AODA algorithm combines the advantages of both timeout-based and dependency-based deadlock analysis approaches, and allows the processes in the timeout state to search for a deadlock circle or knot in the process of dependency transfer. Further, the AODA algorithm cannot lead to false positives and can represent the source of the deadlock accurately. The experimental results on typical MPI communication deadlock benchmarks such as Umpire Test Suit demonstrate the capability of MPIRCDD. Additionally, the experiments on the NPB benchmarks obtain the satisfying performance cost, which show that the MPI-RCDD has strong scalability.

An MPI+OpenACC-Based PRM Scalar Advection Scheme in the GRAPES Model over a Cluster with Multiple CPUs and GPUs

A moisture advection scheme is an essential module of a numerical weather/climate model representing the horizontal transport of water vapor.The Piecewise Rational Method(PRM) scalar advection scheme in the Global/Regional Assimilation and Prediction System(GRAPES) solves the moisture flux advection equation based on PRM.Computation of the scalar advection involves boundary exchange,and computation of higher bandwidth requirements is complicated and time-consuming in GRAPES.Recently,Graphics Processing Units(GPUs) have been widely used to solve scientific and engineering computing problems owing to advancements in GPU hardware and related programming models such as CUDA/OpenCL and Open Accelerator(OpenACC).Herein,we present an accelerated PRM scalar advection scheme with Message Passing Interface(MPI) and OpenACC to fully exploit GPUs’ power over a cluster with multiple Central Processing Units(CPUs) and GPUs,together with optimization of various parameters such as minimizing data transfer,memory coalescing,exposing more parallelism,and overlapping computation with data transfers.Results show that about 3.5 times speedup is obtained for the entire model running at medium resolution with double precision when comparing the scheme’s elapsed time on a node with two GPUs(NVIDIA P100) and two 16-core CPUs(Intel Gold 6142).Further,results obtained from experiments of a higher resolution model with multiple GPUs show excellent scalability.

High Performance MPI over the Slingshot Interconnect

The Slingshot interconnect designed by HPE/Cray is becoming more relevant in high-performance computing with its deployment on the upcoming exascale systems.In particular,it is the interconnect empowering the first exascale and highest-ranked supercomputer in the world,Frontier.It offers various features such as adaptive routing,congestion control,and isolated workloads.The deployment of newer interconnects sparks interest related to performance,scalability,and any potential bottlenecks as they are critical elements contributing to the scalability across nodes on these systems.In this paper,we delve into the challenges the Slingshot interconnect poses with current state-of-the-art MPI(message passing interface)libraries.In particular,we look at the scalability performance when using Slingshot across nodes.We present a comprehensive evaluation using various MPI and communication libraries including Cray MPICH,Open-MPI+UCX,RCCL,and MVAPICH2 on CPUs and GPUs on the Spock system,an early access cluster deployed with Slingshot-10,AMD MI100 GPUs and AMD Epyc Rome CPUs to emulate the Frontier system.We also evaluate preliminary CPU-based support of MPI libraries on the Slingshot-11 interconnect.

High-Performance Flow Classification of Big Data Using Hybrid CPU-GPU Clusters of Cloud Environments

The network switches in the data plane of Software Defined Networking (SDN) are empowered by an elementary process, in which enormous number of packets which resemble big volumes of data are classified into specific flows by matching them against a set of dynamic rules. This basic process accelerates the processing of data, so that instead of processing singular packets repeatedly, corresponding actions are performed on corresponding flows of packets. In this paper, first, we address limitations on a typical packet classification algorithm like Tuple Space Search (TSS). Then, we present a set of different scenarios to parallelize it on different parallel processing platforms, including Graphics Processing Units (GPUs), clusters of Central Processing Units (CPUs), and hybrid clusters. Experimental results show that the hybrid cluster provides the best platform for parallelizing packet classification algorithms, which promises the average throughput rate of 4.2 Million packets per second (Mpps). That is, the hybrid cluster produced by the integration of Compute Unified Device Architecture (CUDA), Message Passing Interface (MPI), and OpenMP programming model could classify 0.24 million packets per second more than the GPU cluster scheme. Such a packet classifier satisfies the required processing speed in the programmable network systems that would be used to communicate big medical data.

GPU acceleration of a nonhydrostatic model for the internal solitary waves simulation

The parallel computing algorithm for a nonhydrostatic model on one or multiple Graphic Processing Units （GPUs） for the simulation of internal solitary waves is presented and discussed. The computational efficiency of the GPU scheme is analyzed by a series of numerical experiments, including an ideal case and the field scale simulations, performed on the workstation and the super- computer system. The calculated results show that the speedup of the developed GPU-based parallel computing scheme, compared to the implementation on a single CPU core, increases with the number of computational grid cells, and the speedup can increase quasi- linearly with respect to the number of involved GPUs for the problem with relatively large number of grid cells within 32 GPUs.

Fast Multicast on Multistage Interconnection Networks Using Multi-Head Worms

This paper proposes a new approach for implementing fast multicast on multistage interconnection networks (MINs) with multi-head worms. For an MIN with n stages of k×k switches, a single multi-head worm can cover an arbitrary set of destinations with a single communication start-up. Compared with schemes using unicast messages, this approach reduces multicast latency significantly and performs better than multi-destination worms.

A new method to retrieve aerosol optical thickness from satellite images on a parallel system

A wide variety of algorithms have been developed to monitor aerosol burden from satellite images. Still, few solutions currently allow for real-time and efficient retrieval of aerosol optical thickness （AOT）, mainly due to the extremely large volume of computation necessary for the numeric solution of atmospheric radiative transfer equations. Taking into account the efforts to exploit the SYNergy of Terra and Aqua Modis （SYNTAM, an AOT retrieval algorithm）, we present in this paper a novel method to retrieve AOT from Moderate Resolution Imaging Spectroradiometer （MODIS） satellite images, in which the strategy of block partition and collective communication was taken, thereby maximizing load balance and reducing the overhead time during inter-processor communication. Experiments were carried out to retrieve AOT at 0.44, 0.55, and 0.67μm of MODIS/Terra and MODIS/Aqua data, using the parallel SYNTAM algorithm in the IBM System Cluster 1600 deployed at China Meteorological Administration （CMA）. Results showed that parallel implementation can greatly reduce computation time, and thus ensure high parallel efficiency. AOT derived by parallel algorithm was validated against measurements from ground-based sun-photometers; in all cases, the relative error range was within 20%, which demonstrated that the parallel algorithm was suitable for applications such as air quality monitoring and climate modeling.

Distributed Heterogeneous Spiking Neural Network Simulator Using Sunway Accelerators

Spiking Neural Network(SNN)simulation is very important for studying brain function and validating the hypotheses for neuroscience,and it can also be used in artificial intelligence.Recently,GPU-based simulators have been developed to support the real-time simulation of SNN.However,these simulators’simulating performance and scale are severely limited,due to the random memory access pattern and the global communication between devices.Therefore,we propose an efficient distributed heterogeneous SNN simulator based on the Sunway accelerators(including SW26010 and SW26010pro),named SWsnn,which supports accurate simulation with small time step(1/16 ms),randomly delay sizes for synapses,and larger scale network computing.Compared with existing GPUs,the Local Dynamic Memory(LDM)(similar to cache)in Sunway is much bigger(4 MB or 16 MB in each core group).To improve the simulation performance,we redesign the network data storage structure and the synaptic plasticity flow to make most random accesses occur in LDM.SWsnn hides Message Passing Interface(MPI)-related operations to reduce communication costs by separating SNN general workflow.Besides,SWsnn relies on parallel Compute Processing Elements(CPEs)rather than serial Manage Processing Element(MPE)to control the communicating buffers,using Register-Level Communication(RLC)and Direct Memory Access(DMA).In addition,SWsnn is further optimized using vectorization and DMA hiding techniques.Experimental results show that SWsnn runs 1.4−2.2 times faster than state-of-the-art GPU-based SNN simulator GPU-enhanced Neuronal Networks(GeNN),and supports much larger scale real-time simulation.