Search NASA⌕ Search

SEARCH · Search NASA

Results for “memory efficiency”

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 937 records · Page 52

MASSIVE HYBRID PARALLELISM FOR FULLY IMPLICIT MULTIPHYSICS

As hardware advances continue to modify the supercomputing landscape, traditional scientific software development practices will become more outdated, ineffective, and inefficient. The process of rewriting/retooling existing software for new architectures is a Sisyphean task, and results in substantial hours of development time, effort, and money. Software libraries which provide an abstraction of the resources provided by such architectures are therefore essential if the computational engineering and science communities are to continue to flourish in this modern computing environment. The Multiphysics Object Oriented Simulation Environment (MOOSE) framework enables complex multiphysics analysis tools to be built rapidly by scientists, engineers, and domain specialists, while also allowing them to both take advantage of current HPC architectures, and efficiently prepare for future supercomputer designs. MOOSE employs a hybrid shared-memory and distributed-memory parallel model and provides a complete and consistent interface for creating multiphysics analysis tools. In this paper, a brief discussion of the mathematical algorithms underlying the framework and the internal object-oriented hybrid parallel design are given. Representative massively parallel results from several applications areas are presented, and a brief discussion of future areas of research for the framework are provided.

97 MATHEMATICS AND COMPUTING↗

Pole EXpansion and Selected Inversion (PEXSI)

The Pole EXpansion and Selected Inversion method (PEXSI) is a fast method for evaluating certain selected elements of a matrix function. PEXSI is highly scalable on distributed memory parallel machines. For sparse matrices, the PEXSI method can be more efficient than the widely used diagonalization method for evaluating matrix functions, especially when a relatively large number of eigenpairs are needed to be computed in the diagonalization methond

Source record↗

Aggressive Sampling for Multi-class to Binary Reduction with Applications to Text Classification

We address the problem of multi-class classification in the case where the number of classes is very large. We propose a double sampling strategy on top of a multi-class to binary reduction strategy, which transforms the original multi-class problem into a binary classification problem over pairs of examples. The aim of the sampling strategy is to overcome the curse of long-tailed class distributions exhibited in majority of large-scale multi-class classification problems and to reduce the number of pairs of examples in the expanded data. We show that this strategy does not alter the consistency of the empirical risk minimization principle defined over the double sample reduction. As a result, experiments are carried out on DMOZ and Wikipedia collections with 10,000 to 100,000 classes where we show the efficiency of the proposed approach in terms of training and prediction time, memory consumption, and predictive performance with respect to state-of-the-art approaches.

97 MATHEMATICS AND COMPUTING↗

Massive hybrid parallelism for fully implicit multiphysics

As hardware advances continue to modify the supercomputing landscape, traditional scientific software development practices will become more outdated, ineffective, and inefficient. The process of rewriting/retooling existing software for new architectures is a Sisyphean task, and results in substantial hours of development time, effort, and money. Software libraries which provide an abstraction of the resources provided by such architectures are therefore essential if the computational engineering and science communities are to continue to flourish in this modern computing environment. The Multiphysics Object Oriented Simulation Environment (MOOSE) framework enables complex multiphysics analysis tools to be built rapidly by scientists, engineers, and domain specialists, while also allowing them to both take advantage of current HPC architectures, and efficiently prepare for future supercomputer designs. MOOSE employs a hybrid shared-memory and distributed-memory parallel model and provides a complete and consistent interface for creating multiphysics analysis tools. In this paper, a brief discussion of the mathematical algorithms underlying the framework and the internal object-oriented hybrid parallel design are given. Representative massively parallel results from several applications areas are presented, and a brief discussion of future areas of research for the framework are provided. (authors)

97 MATHEMATICS AND COMPUTING↗

A new variable-resolution associative memory for high energy physics

We describe an important advancement for the Associative Memory device (AM). The AM is a VLSI processor for pattern recognition based on Content Addressable Memory (CAM) architecture. The AM is optimized for on-line track finding in high-energy physics experiments. Pattern matching is carried out by finding track candidates in coarse resolution 'roads'. A large AM bank stores all trajectories of interest, called 'patterns', for a given detector resolution. The AM extracts roads compatible with a given event during detector read-out. Two important variables characterize the quality of the AM bank: its 'coverage' and the level of fake roads. The coverage, which describes the geometric efficiency of a bank, is defined as the fraction of tracks that match at least one pattern in the bank. Given a certain road size, the coverage of the bank can be increased just adding patterns to the bank, while the number of fakes unfortunately is roughly proportional to the number of patterns in the bank. Moreover, as the luminosity increases, the fake rate increases rapidly because of the increased silicon occupancy. To counter that, we must reduce the width of our roads. If we decrease the road width using the current technology, the system will become very large and extremely expensive. We propose an elegant solution to this problem: the 'variable resolution patterns'. Each pattern and each detector layer within a pattern will be able to use the optimal width, but we will use a 'don't care' feature (inspired from ternary CAMs) to increase the width when that is more appropriate. In other words we can use patterns of variable shape. As a result we reduce the number of fake roads, while keeping the efficiency high and avoiding excessive bank size due to the reduced width. We describe the idea, the implementation in the new AM design and the implementation of the algorithm in the simulation. Finally we show the effectiveness of the 'variable resolution patterns' idea using simulated high occupancy events in the ATLAS detector. (authors)

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Graph Contractions for Calculating Correlation Functions in Lattice QCD

Computing correlation functions for many-particle systems in Lattice QCD is vital to extract nuclear physics observables like the energy spectrum of hadrons such as protons. However, this type of calculation has long been considered to be very challenging and computing-resource intensive because of the complex nature of a hadron composed of quarks with many degrees of freedom. In particular, a correlation function can be calculated through a sum of all possible pairs of quark contractions, each of which is a batched tensor contraction, dictated by Wick's theorem. Because the number of terms of this sum can be very large for any hadronic system of interest, fast evaluation of the sum faces several challenges: an extremely large number of contractions, a huge memory footprint at runtime, and the speed of tensor contractions. In this paper, we present a Lattice QCD analysis software suite, Redstar, which addresses these challenges by utilizing novel algorithmic and software engineering methods targeting modern computing platforms such as many-core CPUs and GPUs. In particular, Redstar represents every term in the sum of a correlation function by a graph, applies efficient graph algorithms to reduce the number of contractions to lower the cost of computations, and minimizes the total memory footprint. Moreover, Redstar carries out the contractions on either CPUs or GPUs utilizing an internal and highly efficient Hadron contraction library. Specifically, we illustrate some important algorithmic optimizations of Redstar, show various key design features of Hadron library, and present the speedup values due to the optimizations along with performance figures for calculating six correlations functions on four computing platforms.

Chen, Jie↗

Graph Contractions for Calculating Correlation Functions in Lattice QCD

Computing correlation functions for many-particle systems in Lattice QCD is vital to extract nuclear physics observables like the energy spectrum of hadrons such as protons. However, this type of calculation has long been considered to be very challenging and computing-resource intensive because of the complex nature of a hadron composed of quarks with many degrees of freedom. In particular, a correlation function can be calculated through a sum of all possible pairs of quark contractions, each of which is a batched tensor contraction, dictated by Wick's theorem. Because the number of terms of this sum can be very large for any hadronic system of interest, fast evaluation of the sum faces several challenges: an extremely large number of contractions, a huge memory footprint at runtime, and the speed of tensor contractions. In this paper, we present a Lattice QCD analysis software suite, Redstar, which addresses these challenges by utilizing novel algorithmic and software engineering methods targeting modern computing platforms such as many-core CPUs and GPUs. In particular, Redstar represents every term in the sum of a correlation function by a graph, applies efficient graph algorithms to reduce the number of contractions to lower the cost of computations, and minimizes the total memory footprint. Moreover, Redstar carries out the contractions on either CPUs or GPUs utilizing an internal and highly efficient Hadron contraction library. Specifically, we illustrate some important algorithmic optimizations of Redstar, show various key design features of Hadron library, and present the speedup values due to the optimizations along with performance figures for calculating six correlations functions on four computing platforms.

Chen, Jie↗

Mini-Ckpts: Surviving OS Failures in Persistent Memory

Concern is growing in the high-performance computing (HPC) community on the reliability of future extreme-scale systems. Current efforts have focused on application fault-tolerance rather than the operating system (OS), despite the fact that recent studies have suggested that failures in OS memory are more likely. The OS is critical to a system's correct and efficient operation of the node and processes it governs -- and in HPC also for any other nodes a parallelized application runs on and communicates with: Any single node failure generally forces all processes of this application to terminate due to tight communication in HPC. Therefore, the OS itself must be capable of tolerating failures. In this work, we introduce mini-ckpts, a framework which enables application survival despite the occurrence of a fatal OS failure or crash. Mini-ckpts achieves this tolerance by ensuring that the critical data describing a process is preserved in persistent memory prior to the failure. Following the failure, the OS is rejuvenated via a warm reboot and the application continues execution effectively making the failure and restart transparent. The mini-ckpts rejuvenation and recovery process is measured to take between three to six seconds and has a failure-free overhead of between 3-5% for a number of key HPC workloads. In contrast to current fault-tolerance methods, this work ensures that the operating and runtime system can continue in the presence of faults. This is a much finer-grained and dynamic method of fault-tolerance than the current, coarse-grained, application-centric methods. Handling faults at this level has the potential to greatly reduce overheads and enables mitigation of additional fault scenarios.

Fiala, David↗

CE TSUNAMI-3D Algorithm Improvements in SCALE 6.2

The TSUNAMI (Tools for Sensitivity and Uncertainty Analysis Methodology Implementation) capabilities within the SCALE code system make use of sensitivity coefficients for an extensive number of criticality safety applications such as quantifying the data-induced uncertainty in the eigenvalue of critical systems, assessing the neutronic similarity between different systems, quantifying computational biases, and guiding nuclear data adjustment studies. The continuous-energy (CE) TSUNAMI-3D code is a new tool included in SCALE 6.2 that allows for eigenvalue and generalized response sensitivity calculations using high-fidelity CE Monte Carlo methods. This paper describes several improvements to CE TSUNAMI-3D beyond those previously published. These improvements primarily consist of - A multi-response capability for the CE TSUNAMI generalized response sensitivity calculation mode; and - Efficiency improvements in CE TSUNAMI's Iterated Fission Probability (IFP)-based algorithms. This paper documents several improvements to the CE TSUNAMI-3D sensitivity algorithms included in the SCALE 6.2 release. These improvements have enabled the simulation of multiple GPT responses within a single CE TSUNAMI simulation, a feature that increases the memory footprint of the simulations by approximately 1-3% per response and the runtime by approximately 1-2%, per response. These algorithmic improvements have also significantly improved the computational efficiency of CE TSUNAMI IFP-based sensitivity calculations, reducing the memory footprints approximately by a factor of 3 and the simulation runtimes by 10-70%. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

A Brief Summary on Formalizing Parallel Tensor Distributions, Redistributions, and Algorithm Derivations

Large-scale datasets in computational chemistry typically require distributed-memory parallel methods to perform a special operation known as tensor contraction. Tensors are multidimensional arrays, and a tensor contraction is akin to matrix multiplication with special types of permutations. Creating an efficient algorithm and optimized implementation in this domain is complex, tedious, and error-prone. To address this, we develop a notation to express data distributions so that we can apply use automated methods to find optimized implementations for tensor contractions. We consider the spin-adapted coupled cluster singles and doubles method from computational chemistry and use our methodology to produce an efficient implementation. Experiments per- formed on the IBM Blue Gene/Q and Cray XC30 demonstrate impact both improved performance and reduced memory consumption.

97 MATHEMATICS AND COMPUTING↗

Magnetism reflectometer study shows LiF layers improve efficiency in spin valve devices

New, more efficient materials for spin valves - a device used in magnetic sensors, random access memories, and hard disk drives - may be on the way based on research using the magnetism reflectometer at Oak Ridge National Laboratory (ORNL). Spin valve devices work by means of two or more conducting magnetic material layers that alternate their electrical resistance depending on the layers alignment. Giant magnetoresistance is a quantum mechanical effect first observed in thin film structures about 20 years ago. The effect is observed as a significant change in electrical resistance, depending on whether the magnetization of adjacent ferromagnetic layers is in a parallel or an antiparallel magnetic alignment. 'What we are doing here is developing new materials. The search for new materials suitable for injecting and transferring carriers with a preferential spin orientation is most important for the development of spintronics,' said Valeria Lauter, lead instrument scientist on the magnetism reflectometer at the Spallation Neutron Source (SNS), who collaborated on the experiment. The researchers discovered that the conductivity of such materials is improved when an organic polymer semiconductor layer is placed between the magnetic materials. Organic semiconductors are now the material of choice for future spin valve devices because they preserve spin coherence over longer times and distances than conventional semiconductors. While research into spin valves has been ongoing, research into organic semiconductors is recent. Previous research has shown that a 'conductivity mismatch' exists in spin valve systems in which ferromagnetic metal electrodes interface with such organic semiconductors as Alq3 ({pi}-conjugated molecule tris(8-hydroxy-quinoline) aluminium). This mismatch limits the efficient injection of the electrons from the electrodes at the interface with the semiconductor material. However, lithium fluoride (LiF), commonly used in light-emitting diodes, has been found to enhance the injection of electrons through the semiconductor. Researchers from the University of Alabama and ORNL used polarized neutrons at the magnetism reflectometer at SNS to investigate the electronic, magnetic, and structural properties of the electrodes in a novel system. In this system, the magnetic layers cobalt and Ni{sub 80}Fe{sub 20} are interfaced with spacer layers composed of the organic semiconductor Alq3. A coupling layer of LiF is inserted to separate the magnetized layers from the semiconductor. 'ALQ3 is an organic semiconductor material,' said Lauter. 'Normally in these systems a first magnetic layer is grown on a hard substrate so that one can get the controlled magnetic parameters. Then you grow the organic semiconductor layer, followed by another magnetic material layer, such as cobalt.' In addition to determining the effect of the LiF layers on the efficiency of the electron injection, the researchers wanted to determine the magnetic properties of the cobalt and Ni{sub 80}Fe{sub 20} as well as the interfacial properties: whether there is interdiffusion of cobalt through the LiF layer to the semiconductor, for example. The researchers used polarized neutrons at beam line 4A to probe the entire, layer-by-layer assembly of the system. 'Reflectometry with polarized neutrons is a perfect method to study thin magnetic films,' Lauter said. 'These thin films - if you put one on a substrate, you see it just like a mirror. However, this mirror has a very complicated internal multilayer structure. The neutrons look inside this complicated structure and characterize each and every interface. Due to the depth sensitivity of the method, we measure the structural and magnetic properties of each layer with the resolution of 0.5 nm. The neutron scattering results found that inserting LiF as a barrier significantly improves the quality of the interface, increasing the injection of electrons from the magnetic layer through the organic semiconductor in the spin valve and enhancing the overall properties of the system. In related work the magnetic properties of the cobalt film and the permalloy Ni{sub 80}Fe{sub 20} were characterized. Cobalt in particular needed attention, as it cannot be grown epitaxially (i.e., deposited) on an organic semiconductor film. Cobalt becomes polycrystalline or amorphous, and this affects its magnetic properties. The data from the first experiment showed that the cobalt layer in the system 'did not have typical magnetic properties,' Lauter said. 'The results showed that the cobalt had low magnetization. To improve the efficiency, the cobalt magnetization should be much higher. So this experiment helped us to improve the growth conditions and to get a cobalt layer with better magnetic properties.' In a subsequent experiment the researchers increased the magnetization of the cobalt, and a follow-up paper is in progress.

36 MATERIALS SCIENCE↗

Efficient Parallelization of a Dynamic Unstructured Application on the Tera MTA

The success of parallel computing in solving real-life computationally-intensive problems relies on their efficient mapping and execution on large-scale multiprocessor architectures. Many important applications are both unstructured and dynamic in nature, making their efficient parallel implementation a daunting task. This paper presents the parallelization of a dynamic unstructured mesh adaptation algorithm using three popular programming paradigms on three leading supercomputers. We examine an MPI message-passing implementation on the Cray T3E and the SGI Origin2OOO, a shared-memory implementation using cache coherent nonuniform memory access (CC-NUMA) of the Origin2OOO, and a multi-threaded version on the newly-released Tera Multi-threaded Architecture (MTA). We compare several critical factors of this parallel code development, including runtime, scalability, programmability, and memory overhead. Our overall results demonstrate that multi-threaded systems offer tremendous potential for quickly and efficiently solving some of the most challenging real-life problems on parallel computers.

Oliker, Leonid↗

The Efficiency and the Scalability of an Explicit Operator on an IBM POWER4 System

We present an evaluation of the efficiency and the scalability of an explicit CFD operator on an IBM POWER4 system. The POWER4 architecture exhibits a common trend in HPC architectures: boosting CPU processing power by increasing the number of functional units, while hiding the latency of memory access by increasing the depth of the memory hierarchy. The overall machine performance depends on the ability of the caches-buses-fabric-memory to feed the functional units with the data to be processed. In this study we evaluate the efficiency and scalability of one explicit CFD operator on an IBM POWER4. This operator performs computations at the points of a Cartesian grid and involves a few dozen floating point numbers and on the order of 100 floating point operations per grid point. The computations in all grid points are independent. Specifically, we estimate the efficiency of the RHS operator (SP of NPB) on a single processor as the observed/peak performance ratio. Then we estimate the scalability of the operator on a single chip (2 CPUs), a single MCM (8 CPUs), 16 CPUs, and the whole machine (32 CPUs). Then we perform the same measurements for a chache-optimized version of the RHS operator. For our measurements we use the HPM (Hardware Performance Monitor) counters available on the POWER4. These counters allow us to analyze the obtained performance results.

Frumkin, Michael↗

Challenges of Future High-End Computing

The next major milestone in high performance computing is a sustained rate of one Pflop/s (also written one petaflops, or 10(circumflex)15 floating-point operations per second). In addition to prodigiously high computational performance, such systems must of necessity feature very large main memories, as well as comparably high I/O bandwidth and huge mass storage facilities. The current consensus of scientists who have studied these issues is that "affordable" petaflops systems may be feasible by the year 2010, assuming that certain key technologies continue to progress at current rates. One important question is whether applications can be structured to perform efficiently on such systems, which are expected to incorporate many thousands of processors and deeply hierarchical memory systems. To answer these questions, advanced performance modeling techniques, including simulation of future architectures and applications, may be required. It may also be necessary to formulate "latency tolerant algorithms" and other completely new algorithmic approaches for certain applications. This talk will give an overview of these challenges.

Bailey, David↗

Scalable All-pairs Shortest Paths for Huge Graphs on Multi-GPU Clusters

We present an optimized Floyd-Warshall (Floyd-Warshall) algorithm that computes the All-pairs shortest path (APSP) for GPU accelerated clusters. The Floyd-Warshall algorithm due to its structural similarities to matrix-multiplication is well suited for highly parallel GPU architectures. To achieve high parallel efficiency, we address two key algorithmic challenges: reducing high communication overhead and addressing limited GPU memory. To reduce high communication costs, we redesign the parallel (a) to expose more parallelism, (b) aggressively overlap communication and computation with pipelined and asynchronous scheduling of operations, and (c) tailored MPI-collective. To cope with limited GPU memory, we employ an offload model, where the data resides on the host and is transferred to GPU on-demand. The proposed optimizations are supported with detailed performance models for tuning. Our optimized parallel Floyd-Warshall implementation is up to 5x faster than a strong baseline and achieves 8.1 PetaFLOPS/sec on 256~nodes of the Summit supercomputer at Oak Ridge National Laboratory. This performance represents 70% of the theoretical peak and 80% parallel efficiency. The offload algorithm can handle 2.5x larger graphs with a 20% increase in overall running time.

Sao, Piyush↗

An enhanced Ada run-time system for real-time embedded processors

An enhanced Ada run-time system has been developed to support real-time embedded processor applications. The primary focus of this development effort has been on the tasking system and the memory management facilities of the run-time system. The tasking system has been extended to support efficient and precise periodic task execution as required for control applications. Event-driven task execution providing a means of task-asynchronous control and communication among Ada tasks is supported in this system. Inter-task control is even provided among tasks distributed on separate physical processors. The memory management system has been enhanced to provide object allocation and protected access support for memory shared between disjoint processors, each of which is executing a distinct Ada program.

Sims, J. T.↗

Concurrent file operations in a high performance FORTRAN

Distributed memory multiprocessor systems can provide the computing power necessary for large scale scientific applications. A critical performance issue for a number of these applications is the efficient transfer of data to secondary storage. Recently several research groups have proposed FORTRAN language extensions for exploiting the data parallelism of such scientific codes on distributed memory architectures. However, few of these high performance FORTRAN's provide appropriate constructs for controlling the use of the parallel I/O capabilities of modern multiprocessing machines. In this paper, we propose constructs to specify I/O operations for distributed data structures in the context of Vienna Fortran. These operations can be used by the programmer to provide information which can help the compiler and runtime environment make the most efficient use of the I/O subsystem.

Brezany, Peter↗

Is Monte Carlo embarrassingly parallel?

Monte Carlo is often stated as being embarrassingly parallel. However, running a Monte Carlo calculation, especially a reactor criticality calculation, in parallel using tens of processors shows a serious limitation in speedup and the execution time may even increase beyond a certain number of processors. In this paper the main causes of the loss of efficiency when using many processors are analyzed using a simple Monte Carlo program for criticality. The basic mechanism for parallel execution is MPI. One of the bottlenecks turn out to be the rendez-vous points in the parallel calculation used for synchronization and exchange of data between processors. This happens at least at the end of each cycle for fission source generation in order to collect the full fission source distribution for the next cycle and to estimate the effective multiplication factor, which is not only part of the requested results, but also input to the next cycle for population control. Basic improvements to overcome this limitation are suggested and tested. Also other time losses in the parallel calculation are identified. Moreover, the threading mechanism, which allows the parallel execution of tasks based on shared memory using OpenMP, is analyzed in detail. Recommendations are given to get the maximum efficiency out of a parallel Monte Carlo calculation. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗