Search NASA⌕ Search

SEARCH · Search NASA

Results for “distributed memory”

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 181 records · Page 10

A high-performance parallel algorithm for nonnegative matrix factorization

Non-negative matrix factorization (NMF) is the problem of determining two non-negative low rank factors W and H, for the given input matrix A, such that A ≈ WH. NMF is a useful tool for many applications in different domains such as topic modeling in text mining, background separation in video analysis, and community detection in social networks. Despite its popularity in the data mining community, there is a lack of efficient distributed algorithms to solve the problem for big data sets. We propose a high-performance distributed-memory parallel algorithm that computes the factorization by iteratively solving alternating non-negative least squares (NLS) subproblems for W and H. It maintains the data and factor matrices in memory (distributed across processors), uses MPI for interprocessor communication, and, in the dense case, provably minimizes communication costs (under mild assumptions). As opposed to previous implementations, our algorithm is also flexible: (1) it performs well for both dense and sparse matrices, and (2) it allows the user to choose any one of the multiple algorithms for solving the updates to low rank factors W and H within the alternating iterations. We demonstrate the scalability of our algorithm and compare it with baseline implementations, showing significant performance improvements.

Kannan, Ramakrishnan↗

HPC-NMF: A High-Performance Parallel Algorithm for Nonnegative Matrix Factorization

NMF is a useful tool for many applications in different domains such as topic modeling in text mining, background separation in video analysis, and community detection in social networks. Despite its popularity in the data mining community, there is a lack of efficient distributed algorithms to solve the problem for big data sets. We propose a high-performance distributed-memory parallel algorithm that computes the factorization by iteratively solving alternating non-negative least squares (NLS) subproblems for $\WW$ and $\HH$. It maintains the data and factor matrices in memory (distributed across processors), uses MPI for interprocessor communication, and, in the dense case, provably minimizes communication costs (under mild assumptions). As opposed to previous implementation, our algorithm is also flexible: It performs well for both dense and sparse matrices, and allows the user to choose any one of the multiple algorithms for solving the updates to low rank factors $\WW$ and $\HH$ within the alternating iterations.

Kannan, Ramakrishnan↗

Computer Sciences and Data Systems, volume 1

Topics addressed include: software engineering; university grants; institutes; concurrent processing; sparse distributed memory; distributed operating systems; intelligent data management processes; expert system for image analysis; fault tolerant software; and architecture research.

Source record↗

Global Arrays Parallel Programming Toolkit

The two predominant classes of programming models for parallel computing are distributed memory and shared memory. Both shared memory and distributed memory models have advantages and shortcomings. Shared memory model is much easier to use but it ignores data locality/placement. Given the hierarchical nature of the memory subsystems in modern computers this characteristic can have a negative impact on performance and scalability. Careful code restructuring to increase data reuse and replacing fine grain load/stores with block access to shared data can address the problem and yield performance for shared memory that is competitive with message-passing. However, this performance comes at the cost of compromising the ease of use that the shared memory model advertises. Distributed memory models, such as message-passing or one-sided communication, offer performance and scalability but they are difficult to program. The Global Arrays toolkit attempts to offer the best features of both models. It implements a shared-memory programming model in which data locality is managed by the programmer. This management is achieved by calls to functions that transfer data between a global address space (a distributed array) and local storage. In this respect, the GA model has similarities to the distributed shared-memory models that provide an explicit acquire/release protocol. However, the GA model acknowledges that remote data is slower to access than local data and allows data locality to be specified by the programmer and hence managed. GA is related to the global address space languages such as UPC, Titanium, and, to a lesser extent, Co-Array Fortran. In addition, by providing a set of data-parallel operations, GA is also related to data-parallel languages such as HPF, ZPL, and Data Parallel C. However, the Global Array programming model is implemented as a library that works with most languages used for technical computing and does not rely on compiler technology for achieving parallel efficiency. It also supports a combination of task- and data-parallelism and is available as an extension of the message passing (MPI) model. The GA model exposes to the programmer the hierarchical memory of modern high-performance computer systems, and by recognizing the communication overhead for remote data transfer, it promotes data reuse and locality of reference. Virtually all the scalable architectures possess non-uniform memory access characteristics that reflect their multi-level memory hierarchies. These hierarchies typically comprise processor registers, multiple levels of cache, local memory, and remote memory. Over time, both the number of levels and the cost (in processor cycles) of accessing deeper levels has been increasing. It is important for any scalable programming model to address memory hierarchy since it is critical to the efficient execution of scalable applications.

ARMCI↗

Contention Modeling for Multithreaded Distributed Shared Memory Machines: The Cray XMT

Distributed Shared Memory (DSM) machines are a wide class of multi-processor computing systems where a large virtually-shared address space is mapped on a network of physically distributed memories. High memory latency and network contention are two of the main factors that limit performance scaling of such architectures. Modern high-performance computing DSM systems have evolved toward exploitation of massive hardware multi-threading and fine-grained memory hashing to tolerate irregular latencies, avoid network hot-spots and enable high scaling. In order to model the performance of such large-scale machines, parallel simulation has been proved to be a promising approach to achieve good accuracy in reasonable times. One of the most critical factors in solving the simulation speed-accuracy trade-off is network modeling. The Cray XMT is a massively multi-threaded supercomputing architecture that belongs to the DSM class, since it implements a globally-shared address space abstraction on top of a physically distributed memory substrate. In this paper, we discuss the development of a contention-aware network model intended to be integrated in a full-system XMT simulator. We start by measuring the effects of network contention in a 128-processor XMT machine and then investigate the trade-off that exists between simulation accuracy and speed, by comparing three network models which operate at different levels of accuracy. The comparison and model validation is performed by executing a string-matching algorithm on the full-system simulator and on the XMT, using three datasets that generate noticeably different contention patterns.

99 GENERAL AND MISCELLANEOUS↗

Modeling the asynchronous Jacobi method without communication delays

Asynchronous iterative methods for solving linear systems are gaining renewed interest due to the high cost of synchronization points in massively parallel codes. Historically, theory on asynchronous iterative methods has focused on asymptotic behavior, while the transient behavior remains poorly understood. In this paper, we study a model of the asynchronous Jacobi method without communication delays, which we call simplified asynchronous Jacobi. Simplified asynchronous Jacobi can be used to model asynchronous Jacobi implemented in shared memory or distributed memory with fast communication networks. Our analysis uses the idea of a propagation matrix, which is similar in concept to an iteration matrix. We show that simplified asynchronous Jacobi can continue to reduce the residual when some processes are slower than other processes. We also show that simplified asynchronous Jacobi can converge when synchronous Jacobi does not. We verify our analysis of simplified asynchronous Jacobi using results from asynchronous Jacobi implemented in shared and distributed memory.

97 MATHEMATICS AND COMPUTING↗

Address tracing for parallel machines

Recently implemented parallel system address-tracing methods based on several metrics are surveyed. The issues specific to collection of traces for both shared and distributed memory parallel computers are highlighted. Five general categories of address-trace collection methods are examined: hardware-captured, interrupt-based, simulation-based, altered microcode-based, and instrumented program-based traces. The problems unique to shared memory and distributed memory multiprocessors are examined separately.

Stunkel, Craig B.↗

Sparse Linear Algebra Toolkit for Computational Aerodynamics

Finding solutions to sparse linear systems of equations is an essential step in Computational Engineering applications of interest to NASA. Linear systems of equations are composed and solved in almost every computational engineering application. The characteristics of linear systems vary greatly from one application to another. Accordingly, there are a wide variety of methods for the solution of linear systems of equations. The operations and methods prepared by the authors are focused on linear systems of interest to NASA, primarily those associated with Computational Fluid Dynamics (CFD), Aeroelasticity, and Aeroacoustics. The Sparse Linear Algebra Toolkit (SLAT) is a coordinated collection of software featuring operations, methods, and data structures that are useful when solving sparse linear systems of equations on modern computer architectures. The implemented operations and methods are designed and tuned for parallelism in shared memory, in distributed memory, and across the hybrid combination of distributed-shared memory. The toolkit includes novel methods and implementations for modern architectures and facilitates development of new approaches for meeting NASA’s evolving computational engineering challenges using evolving computer architectures that are not available in vendor libraries. In this paper, significant features and interfaces within SLAT are presented and verified for simulations performed with NASA’s CFD solver, FUN3D. The runtime and scaling performance of the Generalized Minimum Residual (GMRES) method implemented in SLAT is analyzed for the linear subproblems within the solution of turbulent Navier-Stokes equations employed in the simulation of high-lift configurations. Prior to this work, the SPARSKIT GMRES implementation was the only Krylov subspace method available within FUN3D. A strong scaling study shows the SLAT GMRES implementation facilitates accurate Reynolds-averaged Navier-Stokes CFD solutions between 15% and 56% faster than the SPARSKIT GMRES implementation.

Stephen L Wood↗

Rapid solution of large-scale systems of equations

The analysis and design of complex aerospace structures requires the rapid solution of large systems of linear and nonlinear equations, eigenvalue extraction for buckling, vibration and flutter modes, structural optimization and design sensitivity calculation. Computers with multiple processors and vector capabilities can offer substantial computational advantages over traditional scalar computer for these analyses. These computers fall into two categories: shared memory computers and distributed memory computers. This presentation covers general-purpose, highly efficient algorithms for generation/assembly or element matrices, solution of systems of linear and nonlinear equations, eigenvalue and design sensitivity analysis and optimization. All algorithms are coded in FORTRAN for shared memory computers and many are adapted to distributed memory computers. The capability and numerical performance of these algorithms will be addressed.

Storaasli, Olaf O.↗

Portable parallel stochastic optimization for the design of aeropropulsion components

This report presents the results of Phase 1 research to develop a methodology for performing large-scale Multi-disciplinary Stochastic Optimization (MSO) for the design of aerospace systems ranging from aeropropulsion components to complete aircraft configurations. The current research recognizes that such design optimization problems are computationally expensive, and require the use of either massively parallel or multiple-processor computers. The methodology also recognizes that many operational and performance parameters are uncertain, and that uncertainty must be considered explicitly to achieve optimum performance and cost. The objective of this Phase 1 research was to initialize the development of an MSO methodology that is portable to a wide variety of hardware platforms, while achieving efficient, large-scale parallelism when multiple processors are available. The first effort in the project was a literature review of available computer hardware, as well as review of portable, parallel programming environments. The first effort was to implement the MSO methodology for a problem using the portable parallel programming language, Parallel Virtual Machine (PVM). The third and final effort was to demonstrate the example on a variety of computers, including a distributed-memory multiprocessor, a distributed-memory network of workstations, and a single-processor workstation. Results indicate the MSO methodology can be well-applied towards large-scale aerospace design problems. Nearly perfect linear speedup was demonstrated for computation of optimization sensitivity coefficients on both a 128-node distributed-memory multiprocessor (the Intel iPSC/860) and a network of workstations (speedups of almost 19 times achieved for 20 workstations). Very high parallel efficiencies (75 percent for 31 processors and 60 percent for 50 processors) were also achieved for computation of aerodynamic influence coefficients on the Intel. Finally, the multi-level parallelization strategy that will be needed for large-scale MSO problems was demonstrated to be highly efficient. The same parallel code instructions were used on both platforms, demonstrating portability. There are many applications for which MSO can be applied, including NASA's High-Speed-Civil Transport, and advanced propulsion systems. The use of MSO will reduce design and development time and testing costs dramatically.

Sues, Robert H.↗

A Multi-Level Parallelization Concept for High-Fidelity Multi-Block Solvers

The integration of high-fidelity Computational Fluid Dynamics (CFD) analysis tools with the industrial design process benefits greatly from the robust implementations that are transportable across a wide range of computer architectures. In the present work, a hybrid domain-decomposition and parallelization concept was developed and implemented into the widely-used NASA multi-block Computational Fluid Dynamics (CFD) packages implemented in ENSAERO and OVERFLOW. The new parallel solver concept, PENS (Parallel Euler Navier-Stokes Solver), employs both fine and coarse granularity in data partitioning as well as data coalescing to obtain the desired load-balance characteristics on the available computer platforms. This multi-level parallelism implementation itself introduces no changes to the numerical results, hence the original fidelity of the packages are identically preserved. The present implementation uses the Message Passing Interface (MPI) library for interprocessor message passing and memory accessing. By choosing an appropriate combination of the available partitioning and coalescing capabilities only during the execution stage, the PENS solver becomes adaptable to different computer architectures from shared-memory to distributed-memory platforms with varying degrees of parallelism. The PENS implementation on the IBM SP2 distributed memory environment at the NASA Ames Research Center obtains 85 percent scalable parallel performance using fine-grain partitioning of single-block CFD domains using up to 128 wide computational nodes. Multi-block CFD simulations of complete aircraft simulations achieve 75 percent perfect load-balanced executions using data coalescing and the two levels of parallelism. SGI PowerChallenge, SGI Origin 2000, and a cluster of workstations are the other platforms where the robustness of the implementation is tested. The performance behavior on the other computer platforms with a variety of realistic problems will be included as this on-going study progresses.

Hatay, Ferhat F.↗

Tuning collective communication for Partitioned Global Address Space programming models

Partitioned Global Address Space (PGAS) languages offer programmers the convenience of a shared memory programming style combined with locality control necessary to run on large-scale distributed memory systems. Even within a PGAS language programmers often need to perform global communication operations such as broadcasts or reductions, which are best performed as collective operations in which a group of threads work together to perform the operation. In this study we consider the problem of implementing collective communication within PGAS languages and explore some of the design trade-offs in both the interface and implementation. In particular, PGAS collectives have semantic issues that are different than in send–receive style message passing programs, and different implementation approaches that take advantage of the one-sided communication style in these languages. We present an implementation framework for PGAS collectives as part of the GASNet communication layer, which supports shared memory, distributed memory and hybrids. The framework supports a broad set of algorithms for each collective, over which the implementation may be automatically tuned. In conclusion, we demonstrate the benefit of optimized GASNet collectives using application benchmarks written in UPC, and demonstrate that the GASNet collectives can deliver scalable performance on a variety of state-of-the-art parallel machines including a Cray XT4, an IBM BlueGene/P, and a Sun Constellation system with InfiniBand interconnect.

97 MATHEMATICS AND COMPUTING↗

Relaxing consistency in recoverable distributed shared memory

Relaxed memory consistency models have recently been proposed to tolerate memory access latency in both hardware and software distributed shared memory systems. In recoverable shared memory multiprocessors, relaxing consistency has the added benefit of reducing the number of checkpoints needed to avoid rollback propagation. In this paper, we introduce new checkpointing algorithms that take advantage of relaxed consistency to reduce the performance overhead of checkpointing. We also introduce a scheme based on lazy relaxed consistency, that reduces both checkpointing overhead and the overhead of avoiding error propagation in systems with error latency. Multiprocessor address traces are used to evaluate the relaxed consistency approach to checkpointing with distributed shared memory.

Janssens, Bob↗

Relaxing consistency in recoverable distributed shared memory

Relaxed memory consistency models tolerate increased memory access latency in both hardware and software distributed shared memory systems. In recoverable systems, relaxing consistency has the added benefit of reducing the number of checkpoints needed to avoid rollback propagation. In this paper, we introduce new checkpointing algorithms that take advantage of relaxed consistency to reduce the performance overhead of checkpointing. We also introduce a scheme based on lazy relaxed consistency, that reduces both checkpointing overhead and the overhead of avoiding error propagation in systems with error latency. We use multiprocessor address traces to evaluate the relaxed consistency approach to checkpointing with distributed shared memory.

Janssens, Bob↗

Total recall in distributive associative memories

Iterative error correction of asymptotically large associative memories is equivalent to a one-step learning rule. This rule is the inverse of the activation function of the memory. Spectral representations of nonlinear activation functions are used to obtain the inverse in closed form for Sparse Distributed Memory, Selected-Coordinate Design, and Radial Basis Functions.

Danforth, Douglas G.↗

Per-page control of physical address space distribution among memory modules

Systems, apparatuses, and methods for implementing per-page control of physical address space distribution among memory modules are disclosed. A computing system includes a plurality of processing units coupled to a plurality of memory modules. A determination is made as to which physical address space distribution granularity to implement for physical memory pages allocated for a first data structure. The determination can be made on a per-data-structure basis (e.g., file, page, block, etc.) or on a per-application-basis. A physical address space distribution granularity is encoded as a property of each physical memory page allocated for the first data structure, and physical memory pages of the first data structure distributed across the plurality of memory modules based on a selected physical address space distribution granularity. Page table entries (PTEs) may be annotated with the selected physical address space distribution granularity, using an addressing mapping granularity (AMG) field of a page table entry, where the granularity may be, for example, a fine-grain distribution granularity or a coarse-grain distribution granularity.

Jayasena, Nuwan S.↗

Chapel HyperGraph Library (CHGL)

We present the Chapel Hpergraph Library (CHGL), a library for hypergraph computation in the emerging Chapel language. Hypergraphs generalize graphs, where a hypergraph edge can connect any number of vertices. Thus, hypergraphs capture high-order, high-dimensional interactions between mulitple entities that are not directly expressible in graphs. CHGL is designed to provide HPC-class computation with high-level abstractions and modern language support for parallel comptuing on shared memory and distributed memory systems.

Jenkins, Louis P.↗

Neural networks and MIMD-multiprocessors

Two artificial neural network models are compared. They are the Hopfield Neural Network Model and the Sparse Distributed Memory model. Distributed algorithms for both of them are designed and implemented. The run time characteristics of the algorithms are analyzed theoretically and tested in practice. The storage capacities of the networks are compared. Implementations are done using a distributed multiprocessor system.

Vanhala, Jukka↗