Parallelization strategies for Monte Carlo simulations of thin film deposition

Atomistic simulations of thin film deposition, based on the lattice Monte Carlo method, provide insights into the microstructure evolution at the atomic level. However, large-scale atomistic simulation is limited on a single computer—due to memory and speed constraints. Parallel computation, although promising in memory and speed, has not been widely applied in these simulations because of the intimidating overhead. The key issue in achieving optimal performance is, therefore, to reduce communication overhead among processors. In this paper, we propose a new parallel algorithm for the simulation of large-scale thin film deposition incorporating two optimization strategies: (1) domain decomposition with sub-domain overlapping and (2) asynchronous communication. This algorithm was implemented both on message-passing-processor systems (MPP) and on cluster computers. We found that both architectures are suitable for parallel Monte Carlo simulation of thin film deposition in either a distributed memory mode or a shared memory mode with message-passing libraries.     

基于联合权重超图划分的SNN负载均衡方法

大规模脉冲神经网络并行模拟是探究大脑机能的重要手段。其难点在于合理地将负载映射到并行分布式平台上,提升模拟速度。为解决该问题,提出一种基于联合权重超图划分的SNN负载均衡方法,解决并行计算中进程间计算负载与通信负载的均衡问题,提高SNN模拟速度。并使用稀疏通信的方式替代集体通信,解决事件通信过程中的数据冗余问题,提升通信效率。实验结果表明,该方法使带有STDP突触20%规模的皮质层微电路模型的模拟时间,比标准循环分配算法缩短约64.5%,比普通超图分配算法缩短约57.4%,同时事件通信数据量减少了90%以上。     

Optimization of a thin film deposition process using a dynamic model extracted from molecular simulations

This study presents and demonstrates an algorithm for computing a dynamic model for a thin film deposition process. The proposed algorithm is used on high dimensional Kinetic Monte Carlo (KMC) simulations and consists of applying principal component analysis (PCA) for reducing the state dimension, a self organizing map (SOM) for grouping similar surface configurations and simple cell mapping (SCM) for identifying the transitions between different surface configuration groups. The error associated with this model reduction approach is characterized by running more than 1000 test simulations with highly dynamic and random input profiles. The global error, which is the normalized Euclidean distance between the simulated and predicted states, is found to be less than 1% on average relative to the test simulation results. This indicates that our reduced order dynamic model, which was developed using a rather small simulation set, was able to accurately predict the evolution of the film microstructure for much larger simulation sets and a wide range of process conditions. Minimization of the deposition time to reach a desired film structure has also been achieved using this model. Hence, our study showed that the proposed algorithm is useful for extracting dynamic models from high dimensional and noisy molecular simulation data.     

Domain decomposition methods for the neutron diffusion problem

The neutronic simulation of a nuclear reactor core is performed using the neutron transport equation, and leads to an eigenvalue problem in the steady-state case. Among the deterministic resolution methods, simplified transport (S_PN$S{P}_N$) or diffusion approximations are often used. The MINOS solver developed at CEA Saclay uses a mixed dual finite element method for the resolution of these problems, and has shown his efficiency. In order to take into account the heterogeneities of the geometry, a very fine mesh is generally required, and leads to expensive calculations for industrial applications. In order to take advantage of parallel computers, and to reduce the computing time and the local memory requirement, we propose here two domain decomposition methods based on the MINOS solver. The first approach is a component mode synthesis method on overlapping subdomains: several eigenmodes solutions of a local problem on each subdomain are taken as basis functions used for the resolution of the global problem on the whole domain. The second approach is an iterative method based on a non-overlapping domain decomposition with Robin interface conditions. At each iteration, we solve the problem on each subdomain with the interface conditions given by the solutions on the adjacent subdomains estimated at the previous iteration. Numerical results on parallel computers are presented for the diffusion model on realistic 2D and 3D cores.     

适用于异构集群的混合并行流线生成系统

流线是流场可视化的主要方法之一,而针对大规模流场的流线生成由于计算量大往往需要采用高性能计算机这样的并行计算环境结合并行化算法以实现计算加速.在当前异构计算系统越来越普遍的情况下,为了充分利用并行异构计算环境的计算能力,实现更高效的并行流线生成,本文采用了基于数据并行原语结合分布式消息通讯的技术架构,设计了一套适用于异构集群的混合并行流线生成系统,并在此基础上针对数据分块、数据冗余化及进程通讯策略等方面进行设计,提出并实现了一套并行粒子追踪算法.该系统被部署于国产超算平台上,并针对大规模CFD流场模拟结果数据可视化应用开展了实验.本文给出了相关实验结果,分析了核心并行算法的速度性能、可扩展性以及负载均衡等方面情况,说明了系统及算法的有效性和可扩展性.     

一种基于CUDA的截断重叠维特比译码算法

为解决信道译码在高吞吐量通信系统中的瓶颈问题,通过对CUDA并行计算的了解和对维特比译码并行实现的探索,为卷积码提出了一种基于CUDA的截断重叠维特比译码器。算法通过截断式的子网格图相互重叠的方式,并行执行独立的正向度量计算和回溯过程。实验结果表明,在保证了译码算法误码率性能的同时,获得了良好的吞吐量提升表现,相比现有的实现方式有1.3~3.5倍的提升,降低了硬件开销,能够有效运用于实际高吞吐量通信系统中。     

边缘海静力数值预报模式并行算法研究

边缘海静力数值模式是国内针对边缘海特点自主开发的数值预报模式,但该模式因物理求解方程较多且采用不宜并行化的SOR求解算法而程序计算时间过长。针对上述问题,提出基于三维网格和海洋模式特点的SOR并行求解算法,该算法在保留三维网格数据间依赖关系的同时,有效解决了SOR迭代算法难以并行化的问题。同时,引入通信避免算法,采用MPI非阻塞通信方式,细分计算和通信过程,利用计算有效隐藏通信开销,提高了并行程序效率。实验结果表明,并行后的边缘海静力数值模式程序的性能相对串行程序提升了60.71倍,3天(25920计算时间步)预报结果的均方根误差低于0.001,满足海洋数值预报的时效性和精度要求。     

Visual simulation of GaInP thin film growth

Metal organic chemical vapor deposition (MOCVD) technology is a very efficient way to uniformly grow multi-chip, multilayer and a big area thin film. Kinetic Monte Carlo (KMC) method is one of the important research tools that carry out dynamic simulation of atomic thin films growth. Based on the method of KMC, this paper proposes an algorithm of the process of GaInP thin film grown by MOCVD. KMC simulation and the visualization emulation of GaInP thin film growth in MOCVD reactor are realized. The results of simulation and visualization truly and intuitively displayed process of GaInP thin film growth in MOCVD reactor. The simulation results with this paper’s algorithm well coincide with experimental results. This visualization results provide the optimizations of processing parameters which grow GaInP thin film by MOCVD with theoretical basis.     

基于SNN神经元重分布的NEST仿真器性能优化

为满足大规模脉冲神经网络（SNN）的计算需求,类脑计算系统通常需要采用大规模并行计算平台。然而随着节点数量的增多,通信在仿真中所占比例大幅增加,导致计算效率下降。类脑模拟器开源软件NEST采用缓冲区大小相等的策略,有效缩短了通信时间,但是由于缓冲区互相无交流,使得通信数据量持续增加,因此其在能耗方面表现较差。分析NEST集群的负载特性,针对其中的通信问题进行稀疏性优化,提出基于SNN子图跨节点优化的神经元重分布算法ReLOC。通过优化SNN子图的跨节点分布减少每一轮神经元到进程的数量,从而减少跨节点脉冲,使进程间通信更加稀疏,达到缩减每一轮通信进程的目的。在此基础上,以稀疏交换的思想对NEST本身的通信机制进行改进,使有脉冲交换的进程进行数据交换,从而在连接稀疏的情况下提升通信效率。以包含28个Xilinx PYNQ节点的计算集群作为实验平台,运行皮质微电路SNN模型和平衡随机网络模型,验证ReLOC算法的有效性。实验结果表明,相比循环分布算法,重分布算法能够使通信的平均稀疏性提高20%,同时配合稀疏交换最多可使通信能耗减少98.63%。     

神威太湖之光上分子动力学模拟的性能优化

“神威·太湖之光”国产超级计算机的特点是适用于高通量计算系统,此类系统往往存储器访问延迟,网络延迟较长.在实际应用中,有一大类问题是时间演化的模拟问题,往往需要高频状态迭代,每次迭代需要通信.此类应用问题的典型代表是分子动力学模拟,分子的性质依赖于时间演化,导致状态相关的时间尺度上难以并行化.实际应用中,全原子模型需要模拟超过ms时间尺度,每一步的物理时间为1fs~2.5fs,这意味着所需时间步个数超过10¹²个.众核处理器中,不同核心访存时需较长的“排队”等待,造成访存延迟.另外,网卡通信延迟以及较长的数据通路会带来网络延迟,由此导致在长延迟的众核处理器上进行一次有效的模拟几乎是不可能的.解决此类问题的主要挑战是提高迭代频率,即每秒执行尽可能多的迭代步.针对神威高性能芯片处理器的体系结构特点,以分子动力学模拟为例,研究了一系列优化策略以提高迭代频率：（1）单核通信与片上核间同步相结合,降低通信成本;（2）共享内存等待与从核同步相结合,优化异构体系结构中的核间同步;（3）改变计算模式,减少核间数据关联和依赖关系;（4）数据传输与计算重叠,掩盖访存延迟;（5）规则化问题,以提高访存凝聚性.     

二维宏观拟颗粒并行模拟程序通信性能优化策略

随着粒子模拟并行计算在相关领域应用的不断深入和并行节点计算能力的不断增强,粒子模拟并行程序中通信耗时对整体性能的影响越来越显著,甚至成为主要性能瓶颈.本文在分析影响并行程序通信性能多种因素的基础上,从进程划分方式选择、通信协议优化的角度,对1个典型粒子模拟并行程序——二维宏观拟颗粒并行模拟程序在千兆以太网环境下的通信性能的优化策略进行了测试研究,通过改进并行进程划分方式,采用用户级通信协议等方法,使测试程序通信性能获得明显提高,进而提出了粒子模拟并行程序通信性能优化的思路和建议.     

Parallel Unstructured Dynamic Grid Direct Monte Carlo Simulation of Molecular Gas Dynamics and Its Applications

A new parallel dynamic unstructured grid DSMC method is presented in this paper. The code developed has been applied to the simulation of thin film deposition over microstructures. Surface deformation in such cases poses a challenge for accurate evaluation of gas flow due to the fact that the deposited film thickness is comparable to the feature size. In this study a method is developed to move the mesh at run time. Since in parallel simulation each partition moves independently of the others, a parallel version of moving mesh is proposed to synchronize the displacement of the neighboring partitions, so that there is a smooth transition from one partition to another. An efficient tool for tracking particles during simulation is also presented. Furthermore, the influence of parameters, such as sticking coefficient and aspect ratio on step coverage for a 1m wide trench by sputter deposition was studied. The results showed that the step coverage deteriorated with increasing sticking coefficient and aspect ratio.     

面向异构超算的结构分析高效并行计算方法

In order to exploit the efficient computing power of many integrated cores on heterogeneous cluster, a multi-level and multi-granularity collaborative parallel computing method is proposed for finite element structural mechanical analysis. Computing tasks are divided into three levels: inter-node parallelism, inter-device parallelism and inter-core parallelism. Through mapping decomposablecomput- ing jobs to different hardware layers of heterogeneous MIC system, the proposed method not only effectively resolves the load balancing problem between CPU and MIC devices, but also significantly reduces the communication overheads of the system. Different engineering simulation case experiments for large scale parallel computing were conducted on “Tianhe 2” supercomputer. Up to 39000 CPU+MIC cores were employed and the finite element size of the analysis was more than 100 million units. Test results show that the proposed method can achieve good speedup and parallel computing efficiency in large scale parallel computing of finite element structural analysis. The optimized adaptation of finite element structural analysis and heterogeneous MIC computing platform is realized, which can provide reference for parallel porting and performance optimization of similar applications.     

基于CPU-GPU混合加速的SPH流体仿真方法

基于光滑粒子流体力学SPH的流体仿真是虚拟现实技术的重要研究内容,但SPH流体仿真需要大量的计算资源,采用一般计算方法难以实现流体仿真的实时性。流体仿真通常由物理计算、碰撞检测和渲染等部分组成,借助GPU并行加速粒子的物理属性计算和碰撞过程使SPH方法的实时流体仿真成为可能。为了满足流体仿真应用中的真实性和实时性需求,提出一种基于CPU GPU混合加速的SPH流体仿真方法,流体计算部分采用GPU并行加速,流体渲染部分采用基于CPU的OpenMP加速。实验结果表明,基于CPU GPU混合加速的SPH流体仿真方法与CPU实现相比,能显著地减少流体仿真单帧计算时间且能更快速地完成渲染任务。     

混合重叠网格通信优化算法

重叠网格技术广泛应用在复杂外型和运动边界问题的流场数值模拟中.本文在并行重叠网格隐式挖洞算法实现的基础上,提出了笛卡尔辅助网格和多块结构网格的混合重叠网格方法.通过笛卡尔辅助网格实现重叠网格洞边界和网格插值关系的快速建立.通过定义重叠区域网格权重、部件网格与背景网格绑定的方法,建立了混合网格的并行分配模式,有效减少重叠插值信息在各进程间的通信,实现计算负载和通信负载在各个进程的均匀分配.测试表明该方法可应用于数千万量级的重叠网格系统,可扩展至千核规模,高效的实现多个物体构成的复杂网格系统的重叠关系建立.     

更实际的并行算法的设计

大规模并行计算机的出现和发展迫切要求有新的并行处设计理论和技术来指导更实际的并行算法的设计。本文首先简单介绍了针对ＭＰＣ提出孤ＬｏｇＰ和Ｂａｒｒｉｅｒ－ＬｏｇＰ并行计算模型，然后借助于Ｂａｒｒｉｅｒ－ＬｏｇＰ模型从通信平衡、数据分配和重叠通信与计算这三个方面讨论了更实际的并行算法设计的一般方法和技巧。     

Parallel Monte Carlo simulation in the canonical ensemble on the graphics processing unit

Graphics processing units (GPUs) offer parallel computing power that usually requires a cluster of networked computers or a supercomputer to accomplish. While writing kernel code is fairly straightforward, achieving efficiency and performance requires very careful optimisation decisions and changes to the original serial algorithm. We introduce a parallel canonical ensemble Monte Carlo (MC) simulation that runs entirely on the GPU. In this paper, we describe two MC simulation codes of Lennard-Jones particles in the canonical ensemble, a single CPU core and a parallel GPU implementations. Using Compute Unified Device Architecture, the parallel implementation enables the simulation of systems containing over 200,000 particles in a reasonable amount of time, which allows researchers to obtain more accurate simulation results. A remapping algorithm is introduced to balance the load of the device resources and demonstrate by experimental results that the efficiency of this algorithm is bounded by available GPU resource. Our parallel implementation achieves an improvement of up to 15 times on a commodity GPU over our efficient single core implementation for a system consisting of 256k particles, with the speedup increasing with the problem size. Furthermore, we describe our methods and strategies for optimising our implementation in detail.     

Multi-GPU performance of incompressible flow computation by lattice Boltzmann method on GPU cluster

GPGPU has drawn much attention on accelerating non-graphic applications. The simulation by D3Q19 model of the lattice Boltzmann method was executed successfully on multi-node GPU cluster by using CUDA programming and MPI library. The GPU code runs on the multi-node GPU cluster TSUBAME of Tokyo Institute of Technology, in which a total of 680 GPUs of NVIDIA Tesla are equipped. For multi-GPU computation, domain partitioning method is used to distribute computational load to multiple GPUs and GPU-to-GPU data transfer becomes severe overhead for the total performance. Comparison and analysis were made among the parallel results by 1D, 2D and 3D domain partitionings. As a result, with 384 × 384 × 384 mesh system and 96 GPUs, the performance by 3D partitioning is about 3-4 times higher than that by 1D partitioning. The performance curve is deviated from the idealistic line due to the long communicational time between GPUs. In order to hide the communication time, we introduced the overlapping technique between computation and communication, in which the data transfer process and computation were done in two streams simultaneously. Using 8-96 GPUs, the performances increase by a factor about 1.1-1.3 with a overlapping mode. As a benchmark problem, a large-scaled computation of a flow around a sphere at Re = 13,000 was carried on successfully using the mesh system 2000 × 1000 × 1000 and 100 GPUs. For such a computation with 2 Giga lattice nodes, 6.0 h were used for processing 100,000 time steps. Under this condition, the computational time (2.79 h) and the data communication time (3.06 h) are almost the same.     

A Hybrid Analysis of an Optimization Approach for Cluster Applications

Cluster/distributed computing has become a popular, cost-effective alternative to high-performance parallel computers. Many parallel programming languages and related programming models have become widely accepted on clusters. However, the high communication overhead is a major shortcoming of running parallel applications on cluster/distributed computing environments. To reduce the communication overhead and thus the completion time of a parallel application, this paper introduces and evaluates an efficient Key Message (KM) approach to support parallel computing on cluster computing environments. In this paper, we briefly present the model and algorithm, and then analytical and simulation methods are adopted to evaluate the performance of the algorithm. It demonstrates that when network background load increases or the computation to communication ratio decreases, the analysis results show better improvement on communication of a parallel application over the system which does not use the KM approach.     

大规模问题数据并行性能的分析

从应用的角度建立了评价大规模问题数据并行处理性能的模型,分析了区域的不同划分对解整个问题算法的收敛速度有影响时的并行性能,进而就操作重叠、数据规模以及算法选取等几个方面的问题对大规模数据并行性能所产生的影响进行了分析.最后,给出的例子证明了模型的有效性.