Search NASA⌕ Search

SEARCH · Search NASA

Results for “parallel computer architecture”

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 1,459 records · Page 81

CSB-RNN: A Faster-Than-Realtime RNN Acceleration Framework with Compressed Structured Blocks

Recurrent Neural Networks (RNN) is widely applied to temporal sequence analysis, where real-time performance is usually in demand. However, RNN suffers a heavy computational workload as the model comes with a large weight matrix. To alleviate the pain, model compression (pruning) schemes have been proposed for RNN that pruning the redundant (near-zero) weight-values. On the one hand, the non-structured pruning methods achieve a considerable pruning rate while bringing the computational irregularity, which is un-friendly to parallel-hardware. On the other hand, the existing structured pruning methods consider the hardware parallelism; However, they suffer a poor pruning rate due to the restrict constraints on pruning structure. This paper presents CSB-RNN, an optimized full-stack RNN framework with the novel compressed structured block (CSB) technique. The CSB-pruned RNN model comes with both fine-granularity that benefits the pruning rate and regular structure that facilitates the hardware-parallelism. Further, we propose a novel hardware architecture for inferencing the CSB-pruned model. Different from conventional parallel hardware, this architecture solves the block-workload imbalance issue and achieves an over 95% hardware utilization. With the experiments on 10 RNN models in 5 application domains, the CSB-RNN realizes 7×-20× lossless compression and up to 50× acceptable lossy-compression, which is 2×-7× to the prior art. With the addition of the novel hardware, the compressed-RNN inference reaches a super real-time latency of 10-400µs with FPGA implementation.

Shi, Runbin↗

Automatic Data Traffic Control on DSM Architecture

We study data traffic on distributed shared memory machines and conclude that data placement and grouping improve performance of scientific codes. We present several methods which user can employ to improve data traffic in his code. We report on implementation of a tool which detects the code fragments causing data congestions and advises user on improvements of data routing in these fragments. The capabilities of the tool include deduction of data alignment and affinity from the source code; detection of the code constructs having abnormally high cache or TLB misses; generation of data placement constructs. We demonstrate the capabilities of the tool on experiments with NAS parallel benchmarks and with a simple computational fluid dynamics application ARC3D.

Frumkin, Michael↗

Accelerating science: The usage of commercial clouds in ATLAS Distributed Computing

The ATLAS experiment at CERN is one of the largest scientific machines built to date and will have ever growing computing needs as the Large Hadron Collider collects an increasingly larger volume of data over the next 20 years. ATLAS is conducting R&D projects on Amazon Web Services and Google Cloud as complementary resources for distributed computing, focusing on some of the key features of commercial clouds: lightweight operation, elasticity and availability of multiple chip architectures. The proof of concept phases have concluded with the cloud-native, vendoragnostic integration with the experiment’s data and workload management frameworks. Google Cloud has been used to evaluate elastic batch computing, ramping up ephemeral clusters of up to O(100k) cores to process tasks requiring quick turnaround. Amazon Web Services has been exploited for the successful physics validation of the Athena simulation software on ARM processors. We have also set up an interactive facility for physics analysis allowing endusers to spin up private, on-demand clusters for parallel computing with up to 4 000 cores, or run GPU enabled notebooks and jobs for machine learning applications. The success of the proof of concept phases has led to the extension of the Google Cloud project, where ATLAS will study the total cost of ownership of a production cloud site during 15 months with 10k cores on average, fully integrated with distributed grid computing resources and continue the R&D projects.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Parallel eigenanalysis of finite element models in a completely connected architecture

A parallel algorithm is presented for the solution of the generalized eigenproblem in linear elastic finite element analysis, (K)(phi) = (M)(phi)(omega), where (K) and (M) are of order N, and (omega) is order of q. The concurrent solution of the eigenproblem is based on the multifrontal/modified subspace method and is achieved in a completely connected parallel architecture in which each processor is allowed to communicate with all other processors. The algorithm was successfully implemented on a tightly coupled multiple-instruction multiple-data parallel processing machine, Cray X-MP. A finite element model is divided into m domains each of which is assumed to process n elements. Each domain is then assigned to a processor or to a logical processor (task) if the number of domains exceeds the number of physical processors. The macrotasking library routines are used in mapping each domain to a user task. Computational speed-up and efficiency are used to determine the effectiveness of the algorithm. The effect of the number of domains, the number of degrees-of-freedom located along the global fronts and the dimension of the subspace on the performance of the algorithm are investigated. A parallel finite element dynamic analysis program, p-feda, is documented and the performance of its subroutines in parallel environment is analyzed.

Akl, F. A.↗

Parallel In Situ Coupling of a Simulation with a Fully Featured Visualization System

There is a widening gap between compute performance and the ability to store computation results. Complex scientific codes are the most affected since they must save massive files containing meshes and fields for offline analysis. Time and storage costs instead dictate that data analysis and visualization be combined with the simulations themselves, being done in situ so data are transformed to a manageable size before they are stored. Earlier approaches to in situ processing involved combining specific visualization algorithms into the simulation code, limiting flexibility. We introduce a new library which instead allows a fully-featured visualization tool, VisIt, to request data as needed from the simulation and apply visualization algorithms in situ with minimal modification to the application code.

99 GENERAL AND MISCELLANEOUS↗

Infrastructure for Multiphysics Software Integration in High Performance Computing-Aided Science and Engineering

The project described in this report constructed and exercised an innovative multiphysics coupling toolkit called the Illinois Rocstar MultiPhysics Application Coupling Toolkit (IMPACT). IMPACT is an open source, flexible, natively parallel infrastructure for coupling multiple uniphysics simulation codes into multiphysics computational systems. IMPACT works with codes written in several high-performance-computing (HPC) programming languages, and is designed from the beginning for HPC multiphysics code development. It is designed to be minimally invasive to the individual physics codes being integrated, and has few requirements on those physics codes for integration. The goal of IMPACT is to provide the support needed to enable coupling existing tools together in unique and innovative ways to produce powerful new multiphysics technologies without extensive modification and rewrite of the physics packages being integrated. There are three major outcomes from this project: 1) construction, testing, application, and open-source release of the IMPACT infrastructure, 2) production of example open-source multiphysics tools using IMPACT, and 3) identification and engagement of interested organizations in the tools and applications resulting from the project. This last outcome represents the incipient development of a user community and application echosystem being built using IMPACT. Multiphysics coupling standardization can only come from organizations working together to define needs and processes that span the space of necessary multiphysics outcomes, which Illinois Rocstar plans to continue driving toward. The IMPACT system, including source code, documentation, and test problems are all now available through the public gitHUB.org system to anyone interested in multiphysics code coupling. Many of the basic documents explaining use and architecture of IMPACT are also attached as appendices to this document. Online HTML documentation is available through the gitHUB site. There are over 100 unit tests provided that run through the Illinois Rocstar Application Development (IRAD) lightweight testing infrastructure that is also supplied along with IMPACT. The package as a whole provides an excellent base for developing high-quality multiphysics applications using modern software development practices. To facilitate understanding how to utilize IMPACT effectively, two multiphysics systems have been developed and are available open-source through gitHUB. The simpler of the two systems, named ElmerFoamFSI in the repository, is a multiphysics, fluid-structure-interaction (FSI) coupling of the solid mechanics package Elmer with a fluid dynamics module from OpenFOAM. This coupling illustrates how to combine software packages that are unrelated by either author or architecture and combine them into a robust, parallel multiphysics system. A more complex multiphysics tool is the Illinois Rocstar Rocstar Multiphysics code that was rebuilt during the project around IMPACT. Rocstar Multiphysics was already an HPC multiphysics tool, but now that it has been rearchitected around IMPACT, it can be readily expanded to capture new and different physics in the future. In fact, during this project, the Elmer and OpenFOAM tools were also coupled into Rocstar Multiphysics and demonstrated. The full Rocstar Multiphysics codebase is also available on gitHUB, and licensed for any organization to use as they wish. Finally, the new IMPACT product is already being used in several multiphysics code coupling projects for the Air Force, NASA and the Missile Defense Agency, and initial work on expansion of the IMPACT-enabled Rocstar Multiphysics has begun in support of a commercial company. These initiatives promise to expand the interest and reach of IMPACT and Rocstar Multiphysics, ultimately leading to the envisioned standardization and consortium of users that was one of the goals of this project.

42 ENGINEERING↗

Extending the Roofline Model for Asynchronous Many-Task Runtimes

New architectural trends and extreme scale parallelism challenge the efficient mapping of applications to large scale and emerging systems. Fine-grained Asynchronous Manytask Runtimes (AMR) offer a unique flexibility to hide the varying latencies of contentious operations, and facilitate the effective exploitation of abundant resources. By decomposing work into smaller asynchronous chunks, the impact of unpredictable operations can be mitigated through a tighter overlapping of tasks. Unfortunately, the cost of this decomposition comes at a potentially steep price in the form of runtime overhead which can at times rival the cost of the computation itself. This leads to the question, how large should a task be? A common practice for application developers is to experimentally determine the granularity of a task after a code has been parallelized. Instead, we propose a new methodology based on an extended Roofline model to provide practical upper bounds on the throughput performance of an application. First, we extend the Roofline model to support not only latency hiding analysis, but also a multidimensional amortized analysis. By combining this new methodology with a serial application and an AMR implementation, we can predict the worst case runtime overhead attribution of individual runtime features prior to the development of parallel code. Thus, this runtimecentric methodology can provide a vehicle for application/runtime codesign by providing a comprehensive bottleneck analysis based on existing runtime features.

Suetterlein, Joshua D.↗

Improved algorithms for mapping pipelined and parallel computations

Recent work on the problem of mapping pipelined or parallel computations onto linear array, shared memory, and host-satellite systems is extended. It is shown how these problems can be solved even more efficiently when computation module execution times are bounded from below, intermodule communication times are bounded from above, and the processors satisfy certain homogeneity constraints. The improved algorithms have significantly lower time and space complexities than the more general algorithms: in one case, an O(nm3) time algorithm for mapping m modules onto n processors is replaced with an O(nm log m) time algorithm, and the space requirements are reduced from O(nm2) to O(m). Run-time complexity is reduced further with parallel mapping algorithms based on these improvements, which run on the architectures for which they create mappings.

Nicol, David M.↗

Applying Machine Learning to Understand Write Performance of Large-scale Parallel Filesystems

In high-performance computing (HPC), I/O performance prediction offers the potential to improve the efficiency of scientific computing. In particular, accurate prediction can make runtime estimates more precise, guide users toward optimal checkpoint strategies, and better inform facility provisioning and scheduling policies. HPC I/O performance is notoriously difficult to predict and model, however, in large part because of inherent variability and a lack of transparency in the behaviors of constituent storage system components. In this work we seek to advance the state of the art in HPC I/O performance prediction by (1) modeling the mean performance to address high variability, (2) deriving model features from write patterns, system architecture and system configurations, and (3) employing Lasso regression model to improve model accuracy. We demonstrate the efficacy of our approach by applying it to a crucial subset of common HPC I/O motifs, namely, file-per-process checkpoint write workloads. We conduct experiments on two distinct production HPC platforms—Titan at the Oak Ridge Leadership Computing Facility and Cetus at the Argonne Leadership Computing Facility—to train and evaluate our models. We find that we can attain ≤ 30% relative error for 92.79% and 99.64% of the samples in our test set on these platforms, respectively.

Xie, Bing↗

Feasibility of using the Massively Parallel Processor for large eddy simulations and other Computational Fluid Dynamics applications

The results of an investigation into the feasibility of using the MPP for direct and large eddy simulations of the Navier-Stokes equations is presented. A major part of this study was devoted to the implementation of two of the standard numerical algorithms for CFD. These implementations were not run on the Massively Parallel Processor (MPP) since the machine delivered to NASA Goddard does not have sufficient capacity. Instead, a detailed implementation plan was designed and from these were derived estimates of the time and space requirements of the algorithms on a suitably configured MPP. In addition, other issues related to the practical implementation of these algorithms on an MPP-like architecture were considered; namely, adaptive grid generation, zonal boundary conditions, the table lookup problem, and the software interface. Performance estimates show that the architectural components of the MPP, the Staging Memory and the Array Unit, appear to be well suited to the numerical algorithms of CFD. This combined with the prospect of building a faster and larger MMP-like machine holds the promise of achieving sustained gigaflop rates that are required for the numerical simulations in CFD.

Bruno, John↗

Algorithmic considerations of integrated design for CSI on a hypercube architecture

An approach is presented to the integrated design problem for actively controlled large, flexible mechanical systems for which Control Structure Interaction (CSI) problems are of concern. The two coupled design problems were identified as the optimal Structural Design problem the optimal Controller Design problem. These two problems can be addressed within a decision making loop that would consider each separately, and then sequentially analyze the effects of one on the other. Embedded in such a loop would be the simulation and coordination tasks as part of the decision tools required in a total (software) package. All of the above are compute-intensive tasks. In any such task, possible decompositions and gains due to the inherent parallelism have to be exploited. The problems under consideration, as applied to large flexible mechanical structures, are particularly suited to be mapped onto multicomputer systems in a hypercube topology.

Oezguener, UE.↗

ELSI: A unified software interface for Kohn–Sham electronic structure solvers

Solving the electronic structure from a generalized or standard eigenproblem is often the bottleneck in large scale calculations based on Kohn-Sham density-functional theory. This problem must be addressed by essentially all current electronic structure codes, based on similar matrix expressions, and by high-performance computation. We here present a unified software interface, ELSI, to access different strategies that address the Kohn-Sham eigenvalue problem. Currently supported algorithms include the dense generalized eigensolver library ELPA, the orbital minimization method implemented in libOMM, and the pole expansion and selected inversion (PEXSI) approach with lower computational complexity for semilocal density functionals. The ELSI interface aims to simplify the implementation and optimal use of the different strategies, by offering (a) a unified software framework designed for the electronic structure solvers in Kohn-Sham density-functional theory; (b) reasonable default parameters for a chosen solver; (c) automatic conversion between input and internal working matrix formats, and in the future (d) recommendation of the optimal solver depending on the specific problem. As a result, comparative benchmarks are shown for system sizes up to 11,520 atoms (172,800 basis functions) on distributed memory supercomputing architectures.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Parallelized reliability estimation of reconfigurable computer networks

A parallelized system, ASSURE, for computing the reliability of embedded avionics flight control systems which are able to reconfigure themselves in the event of failure is described. ASSURE accepts a grammar that describes a reliability semi-Markov state-space. From this it creates a parallel program that simultaneously generates and analyzes the state-space, placing upper and lower bounds on the probability of system failure. ASSURE is implemented on a 32-node Intel iPSC/860, and has achieved high processor efficiencies on real problems. Through a combination of improved algorithms, exploitation of parallelism, and use of an advanced microprocessor architecture, ASSURE has reduced the execution time on substantial problems by a factor of one thousand over previous workstation implementations. Furthermore, ASSURE's parallel execution rate on the iPSC/860 is an order of magnitude faster than its serial execution rate on a Cray-2 supercomputer. While dynamic load balancing is necessary for ASSURE's good performance, it is needed only infrequently; the particular method of load balancing used does not substantially affect performance.

Nicol, David M.↗

A parallel algorithm for the eigenvalues and eigenvectors for a general complex matrix

A new parallel Jacobi-like algorithm is developed for computing the eigenvalues of a general complex matrix. Most parallel methods for this parallel typically display only linear convergence. Sequential norm-reducing algorithms also exit and they display quadratic convergence in most cases. The new algorithm is a parallel form of the norm-reducing algorithm due to Eberlein. It is proven that the asymptotic convergence rate of this algorithm is quadratic. Numerical experiments are presented which demonstrate the quadratic convergence of the algorithm and certain situations where the convergence is slow are also identified. The algorithm promises to be very competitive on a variety of parallel architectures.

Shroff, Gautam↗

MarFS, a Near-POSIX Interface to Cloud Objects

The engineering forces driving development of “cloud” storage have produced resilient, cost-effective storage systems that can scale to 100s of petabytes, with good parallel access and bandwidth. These features would make a good match for the vast storage needs of High-Performance Computing datacenters, but cloud storage gains some of its capability from its use of HTTP-style Representational State Transfer (REST) semantics, whereas most large datacenters have legacy applications that rely on POSIX file-system semantics. MarFS is an open-source project at Los Alamos National Laboratory that allows us to present cloud-style object-storage as a scalable near-POSIX file system. We have also developed a new storage architecture to improve bandwidth and scalability beyond what’s available in commodity object stores, while retaining their resilience and economy. Additionally, we present a scheme for scaling the POSIX interface to allow billions of files in a single directory and trillions of files in total.

97 MATHEMATICS AND COMPUTING↗

Application configuration selection for energy-efficient execution on multicore systems

Balanced performance and energy consumption are incorporated in the design of modern computer systems. Several runtime factors, such as concurrency levels, thread mapping strategies, and dynamic voltage and frequency scaling (DVFS) should be considered in order to achieve optimal energy efficiency fora workload. Selecting appropriate run-time factors, however, is one of the most challenging tasks because the run-time factors are architecture-specific and workload-specific. And while most existing works concentrate on either static analysis of the workload or run-time prediction results, we present a hybrid two-step method that utilizes concurrency levels and DVFS settings to achieve the energy efficiency configuration for a worldoad. The experimental results based on a Xeon E5620 server with NPB and PARSEC benchmark suites show that the model is able to predict the energy efficient configuration accurately. On average, an additional 10% EDP (Energy Delay Product) saving is obtained by using run-time DVFS for the entire system. An off-line optimal solution is used to compare with the proposed scheme. Finally, the experimental results show that the average extra EDP saved by the optimal solution is within 5% on selective parallel benchmarks.

97 MATHEMATICS AND COMPUTING↗

Methodology for the Simulation of Molecular Motors at Different Scales

Millisecond-scale conformational transitions represent a seminal challenge for traditional molecular dynamics simulations, even with the help of high-end supercomputer architectures. Such events are particularly relevant to the study of molecular motors—proteins or abiological constructs that convert chemical energy into mechanical work. Here, we present a hybrid-simulation scheme combining an array of methods including elastic network models, transition path sampling, and advanced free-energy methods, possibly in conjunction with generalized-ensemble schemes to deliver a viable option for capturing the millisecond-scale motor steps of biological motors. The methodology is already implemented in large measure in popular molecular dynamics programs, and it can leverage the massively parallel capabilities of petascale computers. The applicability of the hybrid method is demonstrated with two examples, namely cyclodextrin-based motors and V-type ATPases.

Chemistry↗