异构集群中CPU与GPU协同调度算法的设计与实现

为有效提高异构的CPU/GPU集群计算性能,提出一种支持异构集群的CPU与GPU协同计算的两级动态调度算法。根据各节点计算能力评测结果和任务请求动态分发数据,在节点内CPU和GPU之间动态调度任务,使用数据缓存和数据处理双队列机制,提高异构集群的传输和处理效率。该算法实现了集群各节点“能者多劳”,避免了单节点性能瓶颈造成的任务长尾现象。实验结果表明,该算法较传统MPI/GPU并行计算性能提高了11倍。     

基于Linux数据链路层MPI通信机制的设计与实现

针对MPI集群通信的特点,通过分析当前网络的通信结构和MPI的点到点通信模式,提出了一种基于数据链路层的集群通信机制,用以减少协议开销和内存拷贝次数,从而提高集群节点间的通信性能,并且通过实验验证了该机制的可行性。     

基于CPU/GPU异构资源协同调度的改进H-Storm平台

为满足计算密集型大数据应用的实时处理需求,在Apache Storm基础上,研究开发H-Storm异构计算平台。通过多进程服务特性设计图形处理器(GPU)资源的量化和分布式调用机制,进而提出H-Storm异构集群的任务调度策略,实现GPU性能及负载的任务调度算法与协同计算下自适应的流分发决策机制。实验结果表明,在512×512矩阵乘法用例下,与原生Storm平台相比,H-Storm异构计算平台吞吐量提升54.9倍,响应延时下降77倍。     

激光等离子体相互作用模拟的并行和加速研究

随着生成超短激光脉冲技术的不断发展,对这种激光脉冲和等离子体相互作用进行动力学描述也变得越来越重要。PIC(particle-in-cell)是一种在等离子体物理中,研究充能粒子在电磁场中运动轨迹的广泛采用的方法。尽管现在已经有一些在GPU上的PIC方法的实现,但是基于激光等离子体相互作用模拟的特点,仍然有很多重要问题可以尝试其他解决思路。提出了一种把初始的基于CPU的LPI模拟代码完整移植到GPU上的可行方法。提出了一系列加速初始的GPU版本的方法:动态冗余算法、混合精度算法、粒子排序算法。利用并且评估了GPUDirect RDMA(remote direct memory access)技术,其可以提高MPI的通信性能。实验结果证明,与初始的GPU版本相比,"Scatter"阶段加速比为6.1倍,当MPI传输数据大于3 KB时,通信过程提速了2.8倍。这些研究证明了针对模拟应用和GPU集群的特点进行特殊的优化能对性能带来显著的提升。     

基于SMP集群的MPI+CUDA模型的研究与实现

为了研究GPU的通用计算能力和适合SMP集群的编程模型,首次提出MPI+CUDA多粒度混合并行编程的新方法,节点间采用MPI实现粗粒度并行,节点内采用CUDA实现细粒度并行的混合编程方式.利用此方法在搭建的3节点SMP集群环境中,测试了大规模矩阵乘问题的并行计算能力.实验结果表明,该方法能够显著提升并行效率,同时证明MPI+CUDA混合编程模型能够充分发挥SMP集群节点间分布式存储和节点内共享内存的优势,为装有CUDA-enabled GPU的SMP集群提供了一种有效的并行策略.     

面向节点异构GPU集群的编程框架

基于异构GPU集群的主流编程方法是MPI与CUDA的混合编程或者其简单变形。因为对底层的集群架构不透明,程序员对GPU集群采用MPI与CUDA编写应用程序时需要人为考虑硬件计算资源,复杂度高、可移植性差。为此,基于数据流模型设计和实现面向节点异构GPU集群体系结构的新型编程框架分布式并行编程框架(DISPAR)。 DISPAR框架包含2个子系统：(1)代码转换系统StreamCC,是DISPAR源代码到MPI+CUDA代码的自动转换器。(2)任务分配系统StreamMAP,具有自动发现异构计算资源和任务自动映射功能的运行时系统。实验结果表明,该框架有效简化了GPU集群应用程序的编写,可高效地利用异构GPU集群的计算资源,且程序不依赖于硬件平台,可移植性较好。     

MPI全互换通信的性能优化

MPI全互换操作是集群计算机上进行仿真计算时常用的通信操作之一,用于各计算节点间交换上一步骤的中间计算结果。由于全互换通信的密集多对多通信容易产生接收端的阻塞从而增加通信延时,因此通过形成环状的多次规律且有序的通信过程来优化全互换通信操作过程,在大数据量的全互换通信中可以获得明显的性能提升。     

异构GPU集群的任务调度方法研究及实现

GPU集群已经成为高性能计算的重要方式,特别对于计算密集型应用,具有成本低、性能高、功耗小的优势.为了解决GPU集群系统运行中的任务负载均衡问题,文中提出了一种面向计算密集型应用的异构GPU集群调度方法,该方法可以自动发现计算节点,并动态估计计算节点的计算能力,并根据计算能力、任务的计算强度和优先级在异构GPU集群上合理分配计算资源.同时,该系统还具有容错能力,能够处理计算节点的意外退出,可恢复意外退出计算节点的计算任务,并动态适应系统的计算规模.通过实验表明,文中采用的策略达到了预期目的     

异构集群上的宏基因组聚类优化

宏基因组基因聚类是筛选致病基因的新型方法,其依赖于海量的测序数据、有效的聚类算法以及高效的计算机来实现。相关系数矩阵的计算是进行聚类前必须完成的操作,占总计算量的比重较大。以某基因库为例,包含1300个样本、每样本百万基因的数据,单线程运行需要27年。充分发挥多核CPU的潜力,利用GPU加速卡强大的计算能力,将程序扩展到多节点集群上运行,是重要而迫切的工作。在仔细分析算法的基础上,首先针对单CPU节点和单GPU卡做了高效实现,获得了接近理想的加速比;然后利用缓存优化进一步提升性能;最后使用负载均衡方法在MPI线程间分发计算任务,实现了良好的扩展。相比未优化的单线程程序,16节点CPU获得了238.8倍的加速,6 块GPU卡获得了263.8倍的加速。     

分布式深度学习通信架构的性能分析

近年来,深度学习技术的进步推动人工智能进入了一个新的发展时期.但是,海量的训练数据、超大规模的模型给深度学习带来了日益严峻的挑战,分布式深度学习应运而生,逐渐成为应对这一挑战的有效手段,而高效的参数通信架构是保证分布式深度学习性能的关键.针对传统分布式深度学习模型同步架构在大规模节点上并行训练的问题,首先,分析了集中式的Parameter Server和去中心化的Ring Allreduce这2种主流的参数通信架构的原理和性能.然后,在天河高性能GPU集群上基于Ten-sorFlow构建了2种分布式训练架构的对比测试环境.最后,以Parameter Server架构为基准线,测试了Ring Allreduce架构在GPU集群环境下训练AlexNet和ResNet-50的对比性能.实验结果表明,在使用32个GPU的情况下,Ring Allreduce架构扩展效率可达97％,相比Parameter Server架构,其分布式计算性能可提升30％,验证了Ring Allreduce架构具有更好的可扩展性.     

Efficient magnetohydrodynamic simulations on distributed multi-GPU systems using a novel GPU Direct–MPI hybrid approach

Modern graphics processing units (GPUs) have been widely utilized in magnetohydrodynamic (MHD) simulations in recent years. Due to the limited memory of a single GPU, distributed multi-GPU systems are needed to be explored for large-scale MHD simulations. However, the data transfer between GPUs bottlenecks the efficiency of the simulations on such systems. In this paper we propose a novel GPU Direct–MPI hybrid approach to address this problem for overall performance enhancement. Our approach consists of two strategies: (1) We exploit GPU Direct 2.0 to speedup the data transfers between multiple GPUs in a single node and reduce the total number of message passing interface (MPI) communications; (2) We design Compute Unified Device Architecture (CUDA) kernels instead of using memory copy to speedup the fragmented data exchange in the three-dimensional (3D) decomposition. 3D decomposition is usually not preferable for distributed multi-GPU systems due to its low efficiency of the fragmented data exchange. Our approach has made a breakthrough to make 3D decomposition available on distributed multi-GPU systems. As a result, it can reduce the memory usage and computation time of each partition of the computational domain. Experiment results show twice the FLOPS comparing to common 2D decomposition MPI-only implementation method. The proposed approach has been developed in an efficient implementation for MHD simulations on distributed multi-GPU systems, called MGPU–MHD code. The code realizes the GPU parallelization of a total variation diminishing (TVD) algorithm for solving the multidimensional ideal MHD equations, extending our work from single GPU computation (Wong et al., 2011) to multiple GPUs. Numerical tests and performance measurements are conducted on the TSUBAME 2.0 supercomputer at the Tokyo Institute of Technology. Our code achieves 2 TFLOPS in double precision for the problem with 1200³ grid points using 216 GPUs.     

A flexible Patch-based lattice Boltzmann parallelization approach for heterogeneous GPU-CPU clusters

Sustaining a large fraction of single GPU performance in parallel computations is considered to be the major problem of GPU-based clusters. We address this issue in the context of a lattice Boltzmann flow solver that is integrated in the WaLBerla software framework. Our multi-GPU implementation uses a block-structured MPI parallelization and is suitable for load balancing and heterogeneous computations on CPUs and GPUs. The overhead required for multi-GPU simulations is discussed in detail. It is demonstrated that a large fraction of the kernel performance can be sustained for weak scaling on InfiniBand clusters, leading to excellent parallel efficiency. However, in strong scaling scenarios using multiple GPUs is much less efficient than running CPU-only simulations on IBM BG/P and x86-based clusters. Hence, a cost analysis must determine the best course of action for a particular simulation task and hardware configuration. Finally we present weak scaling results of heterogeneous simulations conducted on CPUs and GPUs simultaneously, using clusters equipped with varying node configurations.     

Generating data transfers for distributed GPU parallel programs

Nowadays, high performance applications exploit multiple level architectures, due to the presence of hardware accelerators like GPUs inside each computing node. Data transfers occur at two different levels: inside the computing node between the CPU and the accelerators and between computing nodes. We consider the case where the intra-node parallelism is handled with HMPP compiler directives and message-passing programming with MPI is used to program the inter-node communications. This way of programming on such an heterogeneous architecture is costly and error-prone. In this paper, we specifically demonstrate the transformation of HMPP programs designed to exploit a single computing node equipped with a GPU into an heterogeneous HMPP + MPI exploiting multiple GPUs located on different computing nodes.     

Performance modeling and analysis of heterogeneous lattice Boltzmann simulations on CPU–GPU clusters

Computational fluid dynamic simulations are in general very compute intensive. Only by parallel simulations on modern supercomputers the computational demands of complex simulation tasks can be satisfied. Facing these computational demands GPUs offer high performance, as they provide the high floating point performance and memory to processor chip bandwidth. To successfully utilize GPU clusters for the daily business of a large community, usable software frameworks must be established on these clusters. The development of such software frameworks is only feasible with maintainable software designs that consider performance as a design objective right from the start. For this work we extend the software design concepts to achieve more efficient and highly scalable multi-GPU parallelization within our software framework waLBerla for multi-physics simulations centered around the lattice Boltzmann method. Our software designs now also support a pure-MPI and a hybrid parallelization approach capable of heterogeneous simulations using CPUs and GPUs in parallel. For the first time weak and strong scaling performance results obtained on the Tsubame 2.0 cluster for more than 1000 GPUs are presented using waLBerla. With the help of a new communication model the parallel efficiency of our implementation is investigated and analyzed in a detailed and structured performance analysis. The suitability of the waLBerla framework for production runs on large GPU clusters is demonstrated. As one possible application we show results of strong scaling experiments for flows through a porous medium.     

ChattyGraph:面向异构多协处理器的高可扩展图计算系统

现阶段,随着数据规模扩大化和结构多样化的趋势日益凸现,如何利用现代链路内链的异构多协处理器为大规模数据处理提供实时、可靠的并行运行时环境,已经成为高性能以及数据库领域的研究热点.利用多协处理器(GPU)设备的现代服务器(multi-GPU server)硬件架构环境,已经成为分析大规模、非规则性图数据的首选高性能平台.现有研究工作基于Multi-GPU服务器架构设计的图计算系统和算法(如广度优先遍历和最短路径算法),整体性能已显著优于多核CPU计算环境.然而,这类图计算系统中,多GPU协处理器间的图分块数据传输性能受限于PCI-E总线带宽和局部延迟,导致通过增加GPU设备数量无法达到整体系统性能的类线性增长趋势,甚至会出现严重的时延抖动,进而已无法满足大规模图并行计算系统的高可扩展性要求.经过一系列基准实验验证发现,现有系统存在如下两类缺陷:(1)现代GPU设备间数据通路的硬件架构发展日益更新(如NVLink-V1,NVLink-V2),其链路带宽和延迟得到大幅改进,然而现有系统受限于PCI-E总线进行数据分块通信,无法充分利用现代GPU链路资源(包括链路拓扑、连通性和路由);(2)在...     

Simulation of reaction diffusion processes over biologically relevant size and time scales using multi-GPU workstations

Simulation of in vivo cellular processes with the reaction–diffusion master equation (RDME) is a computationally expensive task. Our previous software enabled simulation of inhomogeneous biochemical systems for small bacteria over long time scales using the MPD-RDME method on a single GPU. Simulations of larger eukaryotic systems exceed the on-board memory capacity of individual GPUs, and long time simulations of modest-sized cells such as yeast are impractical on a single GPU. We present a new multi-GPU parallel implementation of the MPD-RDME method based on a spatial decomposition approach that supports dynamic load balancing for workstations containing GPUs of varying performance and memory capacity. We take advantage of high-performance features of CUDA for peer-to-peer GPU memory transfers and evaluate the performance of our algorithms on state-of-the-art GPU devices. We present parallel efficiency and performance results for simulations using multiple GPUs as system size, particle counts, and number of reactions grow. We also demonstrate multi-GPU performance in simulations of the Min protein system in E. coli. Moreover, our multi-GPU decomposition and load balancing approach can be generalized to other lattice-based problems.     

NUMA-aware image compositing on multi-GPU platform

Sort-last parallel rendering is widely used. Recent GPU developments mean that a PC equipped with multiple GPUs is a viable alternative to a high-cost supercomputer: the Fermi architecture of a single GPU supports uniform virtual addressing, providing a foundation for non-uniform memory access (NUMA) on multi-GPU platforms. Such hardware changes require the user to reconsider the parallel rendering algorithms. In this paper, we thoroughly investigate the NUMA-aware image compositing problem, which is the key final stage in sort-last parallel rendering. Based on a proven radix-k strategy, we find one optimal compositing algorithm, which takes advantage of NUMA architecture on the multi-GPU platform. We quantitatively analyze different image compositing modes for practical image compositing, taking into account peer-to-peer communication costs between GPUs. Our experiments on various datasets show that our image compositing method is very fast, an image of a few megapixels can be composited in about 10 ms by eight GPUs.     

Parallel hyperbolic PDE simulation on clusters: Cell versus GPU

Increasingly, high-performance computing is looking towards data-parallel computational devices to enhance computational performance. Two technologies that have received significant attention are IBM's Cell Processor and NVIDIA's CUDA programming model for graphics processing unit (GPU) computing. In this paper we investigate the acceleration of parallel hyperbolic partial differential equation simulation on structured grids with explicit time integration on clusters with Cell and GPU backends. The message passing interface (MPI) is used for communication between nodes at the coarsest level of parallelism. Optimizations of the simulation code at the several finer levels of parallelism that the data-parallel devices provide are described in terms of data layout, data flow and data-parallel instructions. Optimized Cell and GPU performance are compared with reference code performance on a single x86 central processing unit (CPU) core in single and double precision. We further compare the CPU, Cell and GPU platforms on a chip-to-chip basis, and compare performance on single cluster nodes with two CPUs, two Cell processors or two GPUs in a shared memory configuration (without MPI). We finally compare performance on clusters with 32 CPUs, 32 Cell processors, and 32 GPUs using MPI. Our GPU cluster results use NVIDIA Tesla GPUs with GT200 architecture, but some preliminary results on recently introduced NVIDIA GPUs with the next-generation Fermi architecture are also included. This paper provides computational scientists and engineers who are considering porting their codes to accelerator environments with insight into how structured grid based explicit algorithms can be optimized for clusters with Cell and GPU accelerators. It also provides insight into the speed-up that may be gained on current and future accelerator architectures for this class of applications.

Program summary

Program title: SWsolverCatalogue identifier: AEGY_v1_0Program summary URL:http://cpc.cs.qub.ac.uk/summaries/AEGY_v1_0.htmlProgram obtainable from: CPC Program Library, Queen's University, Belfast, N. IrelandLicensing provisions: GPL v3No. of lines in distributed program, including test data, etc.: 59 168No. of bytes in distributed program, including test data, etc.: 453 409Distribution format: tar.gzProgramming language: C, CUDAComputer: Parallel Computing Clusters. Individual compute nodes may consist of x86 CPU, Cell processor, or x86 CPU with attached NVIDIA GPU accelerator.Operating system: LinuxHas the code been vectorised or parallelized?: Yes. Tested on 1-128 x86 CPU cores, 1-32 Cell Processors, and 1-32 NVIDIA GPUs.RAM: Tested on Problems requiring up to 4 GB per compute node.Classification: 12External routines: MPI, CUDA, IBM Cell SDKNature of problem: MPI-parallel simulation of Shallow Water equations using high-resolution 2D hyperbolic equation solver on regular Cartesian grids for x86 CPU, Cell Processor, and NVIDIA GPU using CUDA.Solution method: SWsolver provides 3 implementations of a high-resolution 2D Shallow Water equation solver on regular Cartesian grids, for CPU, Cell Processor, and NVIDIA GPU. Each implementation uses MPI to divide work across a parallel computing cluster.Additional comments: Sub-program numdiff is used for the test run.     

细粒度任务并行GPU通用矩阵乘

稠密线性代数运算对模式识别和生物信息等许多实际应用至关重要,而通用矩阵乘(GEMM)处于稠密线性代数运算的基础地位。在cuBLAS与MAGMA中,GEMM被实现为若干kernel函数,对大型GEMM计算能够达到很高的性能。然而,现有实现对批量的小型GEMM计算性能发挥则较为有限。而且,现有实现也不能在多个具有不同性能的GPU之间自动扩展并达到负载均衡。提出任务并行式GEMM(TPGEMM),用细粒度任务并行的方式实现批量矩阵乘和多GPU矩阵乘。一个或多个GEMM的计算能够被拆分为多个任务,动态地调度到一个或多个GPU上。TPGEMM避免了为批量矩阵乘启动多个kernel函数的开销,对批量矩阵乘能够取得显著高于cuBLAS与MAGMA的性能。在低开销细粒度任务调度的基础上,TPGEMM支持单个GEMM计算在多个GPU间的自动并行,在一台具有四个不同性能GPU的工作站上取得了接近100%的扩展效率。     

Accelerating incompressible flow computations with a Pthreads-CUDA implementation on small-footprint multi-GPU platforms

Graphics processor units (GPU) that are originally designed for graphics rendering have emerged as massively-parallel “co-processors” to the central processing unit (CPU). Small-footprint multi-GPU workstations with hundreds of processing elements can accelerate compute-intensive simulation science applications substantially. In this study, we describe the implementation of an incompressible flow Navier–Stokes solver for multi-GPU workstation platforms. A shared-memory parallel code with identical numerical methods is also developed for multi-core CPUs to provide a fair comparison between CPUs and GPUs. Specifically, we adopt NVIDIA’s Compute Unified Device Architecture (CUDA) programming model to implement the discretized form of the governing equations on a single GPU. Pthreads are then used to enable communication across multiple GPUs on a workstation. We use separate CUDA kernels to implement the projection algorithm to solve the incompressible fluid flow equations. Kernels are implemented on different memory spaces on the GPU depending on their arithmetic intensity. The memory hierarchy specific implementation produces significantly faster performance. We present a systematic analysis of speedup and scaling using two generations of NVIDIA GPU architectures and provide a comparison of single and double precision computational performance on the GPU. Using a quad-GPU platform for single precision computations, we observe two orders of magnitude speedup relative to a serial CPU implementation. Our results demonstrate that multi-GPU workstations can serve as a cost-effective small-footprint parallel computing platform to accelerate computational fluid dynamics (CFD) simulations substantially.