Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “distributed memory parallel algorithm”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 217 records · Page 12

Cross-scale Efficient Tensor Contractions for Coupled Cluster Computations Through Multiple Programming Model Backends

Coupled-cluster methods provide highly accurate models of molecular structure by explicit numerical calculation of tensors representing the correlation between electrons. These calculations are dominated by a sequence of tensor contractions, motivating the development of numerical libraries for such operations. While based on matrix-matrix multiplication, these libraries are specialized to exploit symmetries in the molecular structure and in electronic interactions, and thus reduce the size of the tensor representation and the complexity of contractions. The resulting algorithms are irregular and their parallelization has been previously achieved via the use of dynamic scheduling or specialized data decompositions. We introduce our efforts to extend the Libtensor framework to work in the distributed memory environment in a scalable and energy efficient manner. We achieve up to 240 speedup compared with the best optimized shared memory implementation. We attain scalability to hundreds of thousands of compute cores on three distributed-memory architectures, (Cray XC30&XC40, BlueGene/Q), and on a heterogeneous GPU-CPU system (Cray XK7). As the bottlenecks shift from being compute-bound DGEMM's to communication-bound collectives as the size of the molecular system scales, we adopt two radically different parallelization approaches for handling load-imbalance. Nevertheless, we preserve a uni ed interface to both programming models to maintain the productivity of computational quantum chemists.

74 ATOMIC AND MOLECULAR PHYSICS↗

Portability and Cross-Platform Performance of an MPI-Based Parallel Polygon Renderer

Visualizing the results of computations performed on large-scale parallel computers is a challenging problem, due to the size of the datasets involved. One approach is to perform the visualization and graphics operations in place, exploiting the available parallelism to obtain the necessary rendering performance. Over the past several years, we have been developing algorithms and software to support visualization applications on NASA's parallel supercomputers. Our results have been incorporated into a parallel polygon rendering system called PGL. PGL was initially developed on tightly-coupled distributed-memory message-passing systems, including Intel's iPSC/860 and Paragon, and IBM's SP2. Over the past year, we have ported it to a variety of additional platforms, including the HP Exemplar, SGI Origin2OOO, Cray T3E, and clusters of Sun workstations. In implementing PGL, we have had two primary goals: cross-platform portability and high performance. Portability is important because (1) our manpower resources are limited, making it difficult to develop and maintain multiple versions of the code, and (2) NASA's complement of parallel computing platforms is diverse and subject to frequent change. Performance is important in delivering adequate rendering rates for complex scenes and ensuring that parallel computing resources are used effectively. Unfortunately, these two goals are often at odds. In this paper we report on our experiences with portability and performance of the PGL polygon renderer across a range of parallel computing platforms.

Crockett, Thomas W.↗

Locality-Aware Parallel Process Mapping for Multi-Core HPC Systems

High Performance Computing (HPC) systems are composed of servers containing an ever-increasing number of cores. With such high processor core counts, non-uniform memory access (NUMA) architectures are almost universally used to reduce inter-processor and memory communication bottlenecks by distributing processors and memory throughout a server-internal networking topology. Application studies have shown that the tuning of processes placement in a server s NUMA networking topology to the application can have a dramatic impact on performance. The performance implications are magnified when running a parallel job across multiple server nodes, especially with large scale HPC applications. This paper presents the Locality-Aware Mapping Algorithm (LAMA) for distributing the individual processes of a parallel application across processing resources in an HPC system, paying particular attention to the internal server NUMA topologies. The algorithm is able to support both homogeneous and heterogeneous hardware systems, and dynamically adapts to the available hardware and user-specified process layout at run-time. As implemented in Open MPI, the LAMA provides 362,880 mapping permutations and is able to naturally scale out to additional hardware resources as they become available in future architectures.

99 GENERAL AND MISCELLANEOUS↗

Livermore Big Artificial Neural Network Toolkit

LBANN is a toolkit that is designed to train artificial neural networks efficiently on high performance computing architectures. It is optimized to take advantages of key High Performance Computing features to accelerate neural network training. Specifically it is optimized for low-latency, high bandwidth interconnects, node-local NVRAM, node-local GPU accelerators, and high bandwidth parallel file systems. It is built on top of the open source Elemental distributed-memory dense and spars-direct linear algebra and optimization library that is released under the BSD license. The algorithms contained within LBANN are drawn from the academic literature and implemented to work within a distributed-memory framework.

Essen, Brian Van↗

Performance Assessment of OVERFLOW on Distributed Computing Environment

The aerodynamic computer code, OVERFLOW, with a multi-zone overset grid feature, has been parallelized to enhance its performance on distributed and shared memory paradigms. Practical application benchmarks have been set to assess the efficiency of code's parallelism on high-performance architectures. The code's performance has also been experimented with in the context of the distributed computing paradigm on distant computer resources using the Information Power Grid (IPG) toolkit, Globus. Two parallel versions of the code, namely OVERFLOW-MPI and -MLP, have developed around the natural coarse grained parallelism inherent in a multi-zonal domain decomposition paradigm. The algorithm invokes a strategy that forms a number of groups, each consisting of a zone, a cluster of zones and/or a partition of a large zone. Each group can be thought of as a process with one or multithreads assigned to it and that all groups run in parallel. The -MPI version of the code uses explicit message-passing based on the standard MPI library for sending and receiving interzonal boundary data across processors. The -MLP version employs no message-passing paradigm; the boundary data is transferred through the shared memory. The -MPI code is suited for both distributed and shared memory architectures, while the -MLP code can only be used on shared memory platforms. The IPG applications are implemented by the -MPI code using the Globus toolkit. While a computational task is distributed across multiple computer resources, the parallelism can be explored on each resource alone. Performance studies are achieved with some practical aerodynamic problems with complex geometries, consisting of 2.5 up to 33 million grid points and a large number of zonal blocks. The computations were executed primarily on SGI Origin 2000 multiprocessors and on the Cray T3E. OVERFLOW's IPG applications are carried out on NASA homogeneous metacomputing machines located at three sites, Ames, Langley and Glenn. Plans for the future will exploit the distributed parallel computing capability on various homogeneous and heterogeneous resources and large scale benchmarks. Alternative IPG toolkits will be used along with sophisticated zonal grouping strategies to minimize the communication time across the computer resources.

Djomehri, M. Jahed↗

Distributed non-equilibrium Green’s function algorithms for the simulation of nanoelectronic devices with scattering

Through the non-equilibrium Green’s function (NEGF) formalism, quantum-scale device simulation can be performed with the inclusion of electron-phonon scattering. Yet, the simulation of realistically sized devices under the NEGF formalism typically requires prohibitive amounts of memory and computation time. Two of the most demanding computational problems for NEGF simulation involve mathematical operations with structured matrices called semiseparable matrices. In this work, we present parallel approaches for these computational problems which allow for efficient distribution of both memory and computation based upon the underlying device structure. This is vital when simulating realistically sized devices due to the aforementioned computational burdens. First, we consider determining a distributed compact representation for the retarded Green’s function matrix G R . This compact representation is exact and allows for any entry in the matrix to be generated through the inherent semiseparable structure. The second parallel operation allows for the computation of electron density and current characteristics for the device. Specifically, matrix products between the distributed representation for the semiseparable matrix G R and the self-energy scattering terms in Σ < produce the less-than Green’s function G < . As an illustration of the computational efficiency of our approach, we stably generate the mobility for nanowires with cross-sectional sizes as large as 4.5 nm, assuming an atomistic model with scattering.

47 OTHER INSTRUMENTATION↗

Integrated Model, Batch and Domain Parallelism in Training Neural Networks

We propose a new integrated method of exploiting model, batch and domain parallelism for the training of deep neural networks (DNNs) on large distributed-memory computers using minibatch stochastic gradient descent (SGD). Our goal is to find an efficient parallelization strategy for a fixed batch size using $P$ processes. Our method is inspired by the communication-avoiding algorithms in numerical linear algebra. We see $P$ processes as logically divided into a $P_r \times P_c$ grid where the $P_r$ dimension is implicitly responsible for model/domain parallelism and the $P_c$ dimension is implicitly responsible for batch parallelism. In practice, the integrated matrix-based parallel algorithm encapsulates these types of parallelism automatically. We analyze the communication complexity and analytically demonstrate that the lowest communication costs are often achieved neither with pure model nor with pure data parallelism. We also show how the domain parallel approach can help in extending the theoretical scaling limit of the typical batch parallel method.

97 MATHEMATICS AND COMPUTING↗

Parallel phase model : a programming model for high-end parallel machines with manycores.

This paper presents a parallel programming model, Parallel Phase Model (PPM), for next-generation high-end parallel machines based on a distributed memory architecture consisting of a networked cluster of nodes with a large number of cores on each node. PPM has a unified high-level programming abstraction that facilitates the design and implementation of parallel algorithms to exploit both the parallelism of the many cores and the parallelism at the cluster level. The programming abstraction will be suitable for expressing both fine-grained and coarse-grained parallelism. It includes a few high-level parallel programming language constructs that can be added as an extension to an existing (sequential or parallel) programming language such as C; and the implementation of PPM also includes a light-weight runtime library that runs on top of an existing network communication software layer (e.g. MPI). Design philosophy of PPM and details of the programming abstraction are also presented. Several unstructured applications that inherently require high-volume random fine-grained data accesses have been implemented in PPM with very promising results.

99 GENERAL AND MISCELLANEOUS↗

Modular Spiking Neural Circuits for Mapping Long Short-Term Memory on a Neurosynaptic Processor

Due to the distributed and asynchronous nature of neural computation through low-energy spikes, brain-inspired hardware systems offer high energy efficiency and massive parallelism. One such platform is the IBM TrueNorth neurosynaptic system. Recently, TrueNorth compatible representation learning algorithms have emerged, achieving close to the state-of-the-art performance in various data sets. However, its application in temporal sequence processing models, such as recurrent neural networks (RNNs), is still only at the proof of concept level. There is an inherent difficulty in capturing temporal dynamics of an RNN using spiking neurons, which is only exasperated by the hardware constraints in connectivity and synaptic weight resolution. This paper presents a design flow that overcomes these difficulties and maps a special case of recurrent networks called long short-term memory (LSTM) onto a spike-based platform. The framework utilizes various approximation techniques, such as activation discretization, weight quantization, and scaling and rounding, spiking neural circuits that implement the complex gating mechanisms, and a store-and-release technique to enable neuron synchronization and faithful storage. While the presented techniques can be applied to map LSTM to any spiking neural network (SNN) simulator/emulator, here we choose the TrueNorth chip as the target platform by adhering to its hardware constraints. Three LSTM applications, parity check, extended Reber grammar, and question classification, are evaluated. The tradeoffs among accuracy, performance, and energy tradeoffs achieved on TrueNorth are demonstrated. Finally, this is compared with the performance on an SNN platform without hardware constraints, which represents the upper bound of the achievable accuracy.

97 MATHEMATICS AND COMPUTING↗

Scalable Computation of Topological Abstractions for Scalar Data

Topological data analysis has become an important tool for large scale scalar data analysis and visualization, efficiently extracting the inherent structure and features of interest of the data. However, with growing dataset sizes and complexity, it is increasingly becoming infeasible to compute topological abstractions of interest in serial and on single machines. This paper presents the state of the art in the scalable computation of topological abstractions on scalar data, in shared memory parallel on single machines, and in distributed memory parallel on multiple machines. We highlight results for set‐based, graph‐based and complex‐based abstractions and organize the state of the art based on this taxonomy. The paper identifies parallelization and distribution techniques common in topological algorithms and highlights further areas of interest with underdeveloped efforts.

97 MATHEMATICS AND COMPUTING↗

A Comparison of PETSC Library and HPF Implementations of an Archetypal PDE Computation

Two paradigms for distributed-memory parallel computation that free the application programmer from the details of message passing are compared for an archetypal structured scientific computation a nonlinear, structured-grid partial differential equation boundary value problem using the same algorithm on the same hardware. Both paradigms, parallel libraries represented by Argonne's PETSC, and parallel languages represented by the Portland Group's HPF, are found to be easy to use for this problem class, and both are reasonably effective in exploiting concurrency after a short learning curve. The level of involvement required by the application programmer under either paradigm includes specification of the data partitioning (corresponding to a geometrically simple decomposition of the domain of the PDE). Programming in SPAM style for the PETSC library requires writing the routines that discretize the PDE and its Jacobian, managing subdomain-to-processor mappings (affine global- to-local index mappings), and interfacing to library solver routines. Programming for HPF requires a complete sequential implementation of the same algorithm, introducing concurrency through subdomain blocking (an effort similar to the index mapping), and modest experimentation with rewriting loops to elucidate to the compiler the latent concurrency. Correctness and scalability are cross-validated on up to 32 nodes of an IBM SP2.

Hayder, M. Ehtesham↗

A multiarchitecture parallel-processing development environment

A description is given of the hardware and software of a multiprocessor test bed - the second generation Hypercluster system. The Hypercluster architecture consists of a standard hypercube distributed-memory topology, with multiprocessor shared-memory nodes. By using standard, off-the-shelf hardware, the system can be upgraded to use rapidly improving computer technology. The Hypercluster's multiarchitecture nature makes it suitable for researching parallel algorithms in computational field simulation applications (e.g., computational fluid dynamics). The dedicated test-bed environment of the Hypercluster and its custom-built software allows experiments with various parallel-processing concepts such as message passing algorithms, debugging tools, and computational 'steering'. Such research would be difficult, if not impossible, to achieve on shared, commercial systems.

Townsend, Scott↗

Distributed Halide

Many image processing tasks are naturally expressed as a pipeline of small computational kernels known as stencils. Halide is a popular domain-specific language and compiler designed to implement image processing algorithms. Halide uses simple language constructs to express what to compute and a separate scheduling co-language for expressing when and where to perform the computation. This approach has demonstrated performance comparable to or better than hand-optimized code. Until now, however, Halide has been restricted to parallel shared memory execution, limiting its performance for memory-bandwidth-bound pipelines or large-scale image processing tasks. We present an extension to Halide to support distributed-memory parallel execution of complex stencil pipelines. These extensions compose with the existing scheduling constructs in Halide, allowing expression of complex computation and communication strategies. Existing Halide applications can be distributed with minimal changes, allowing programmers to explore the tradeoff between recomputation and communication with little effort. Approximately 10 new of lines code are needed even for a 200 line, 99 stage application. On nine image processing benchmarks, our extensions give up to a 1.4× speedup on a single node over regular multithreaded execution with the same number of cores, by mitigating the effects of non-uniform memory access. The distributed benchmarks achieve up to 18× speedup on a 16 node testing machine and up to 57× speedup on 64 nodes of the NERSC Cori supercomputer.

97 MATHEMATICS AND COMPUTING↗

Streaming data analytics via message passing with application to graph algorithms

The need to process streaming data, which arrives continuously at high-volume in real-time, arises in a variety of contexts including data produced by experiments, collections of environmental or network sensors, and running simulations. Streaming data can also be formulated as queries or transactions which operate on a large dynamic data store, e.g. a distributed database. We describe a lightweight, portable framework named PHISH which enables a set of independent processes to compute on a stream of data in a distributed-memory parallel manner. Datums are routed between processes in patterns defined by the application. PHISH can run on top of either message-passing via MPI or sockets via ZMQ. The former means streaming computations can be run on any parallel machine which supports MPI; the latter allows them to run on a heterogeneous, geographically dispersed network of machines. We illustrate how PHISH can support streaming MapReduce operations, and describe streaming versions of three algorithms for large, sparse graph analytics: triangle enumeration, subgraph isomorphism matching, and connected component finding. Lastly, we also provide benchmark timings for MPI versus socket performance of several kernel operations useful in streaming algorithms.

97 MATHEMATICS AND COMPUTING↗

Aho-Corasick String Matching on Shared and Distributed Memory Parallel Architectures

String matching is at the core of many critical applications, including network intrusion detection systems, search engines, virus scanners, spam filters, DNA and protein sequencing, and data mining. For all of these applications string matching requires a combination of (sometimes all) the following characteristics: high and/or predictable performance, support for large data sets and flexibility of integration and customization. Many software based implementations targeting conventional cache-based microprocessors fail to achieve high and predictable performance requirements, while Field-Programmable Gate Array (FPGA) implementations and dedicated hardware solutions fail to support large data sets (dictionary sizes) and are difficult to integrate and customize. The advent of multicore, multithreaded, and GPU-based systems is opening the possibility for software based solutions to reach very high performance at a sustained rate. This paper compares several software-based implementations of the Aho-Corasick string searching algorithm for high performance systems. We discuss the implementation of the algorithm on several types of shared-memory high-performance architectures (Niagara 2, large x86 SMPs and Cray XMT), distributed memory with homogeneous processing elements (InfiniBand cluster of x86 multicores) and heterogeneous processing elements (InfiniBand cluster of x86 multicores with NVIDIA Tesla C10 GPUs). We describe in detail how each solution achieves the objectives of supporting large dictionaries, sustaining high performance, and enabling customization and flexibility using various data sets.

99 GENERAL AND MISCELLANEOUS↗

A Parallel EM Algorithm for Model-Based Clustering Applied to the Exploration of Large Spatio-Temporal Data

We develop a parallel EM algorithm for multivariate Gaussian mixture models and use it to perform model-based clustering of a large climate data set. Three variants of the EM algorithm are reformulated in parallel and a new variant that is faster is presented. All are implemented using the single program, multiple data (SPMD) programming model, which is able to take advantage of the combined collective memory of large distributed computer architectures to process larger data sets. Displays of the estimated mixture model rather than the data allow us to explore multivariate relationships in a way that scales to arbitrary size data. We study the performance of our methodology on simulated data and apply our methodology to a high resolution climate dataset produced by the community atmosphere model (CAM5). This article has supplementary material online.

Chen, Wei-Chen↗

MICADO: Parallel implementation of a 2D-1D iterative algorithm for the 3D neutron transport problem in prismatic geometries

The large increase in computing power over the past few years now makes it possible to consider developing 3D full-core heterogeneous deterministic neutron transport solvers for reference calculations. Among all approaches presented in the literature, the method first introduced in [1] seems very promising. It consists in iterating over resolutions of 2D and ID MOC problems by taking advantage of prismatic geometries without introducing approximations of a low order operator such as diffusion. However, before developing a solver with all industrial options at EDF, several points needed to be clarified. In this work, we first prove the convergence of this iterative process, under some assumptions. We then present our high-performance, parallel implementation of this algorithm in the MICADO solver. Benchmarking the solver against the Takeda case shows that the 2D-1D coupling algorithm does not seem to affect the spatial convergence order of the MOC solver. As for performance issues, our study shows that even though the data distribution is suited to the 2D solver part, the efficiency of the ID part is sufficient to ensure a good parallel efficiency of the global algorithm. After this study, the main remaining difficulty implementation-wise is about the memory requirement of a vector used for initialization. An efficient acceleration operator will also need to be developed. (authors)

97 MATHEMATICS AND COMPUTING↗

A balanced submatrix merging algorithm for multiprocessor architectures

In this article, a parallel algorithm which applies Givens rotations to selectively annihilate k(k + 1)/2 nonzero elements from two k x n(k not more than n) upper trapezoidal submatrices is described. The new algorithm is suitable for implementation on either a pair of directly connected local-memory processors or two clusters of multiple tightly-coupled processors. Analyses show that in both cases the proposed algorithms achieve optimal speed-up by balancing the work load distribution and masking interprocessor or intercluster communication by computation if k is much small than n. In the context of solving large scale least squares problems, this submatrix merging step is repetitively needed during the entire computation and, furthermore, there are usually many pairs of such submatrices to be merged with each submatrix stored in the memory of a processor or a cluster of processors. The proposed algorithm can be applied to each pair of submatrices concurrently, and thus parallelizes an important step in solving the least squares problems.

Chu, Eleanor↗