Search NASA⌕ Search

SEARCH · Search NASA

Results for “Parallel processing”

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 1,081 records · Page 60

Communication Characterization and Optimization of Applications Using Topology-Aware Task Mapping on Large Supercomputers

On large supercomputers, the job scheduling systems may assign a non-contiguous node allocation for user applications depending on available resources. With parallel applications using MPI (Message Passing Interface), the default process ordering does not take into account the actual physical node layout available to the application. This contributes to non-locality in terms of physical network topology and impacts communication performance of the application. In order to mitigate such performance penalties, this work describes techniques to identify suitable task mapping that takes the layout of the allocated nodes as well as the application's communication behavior into account. During the first phase of this research, we instrumented and collected performance data to characterize communication behavior of critical US DOE (United States - Department of Energy) applications using an augmented version of the mpiP tool. Subsequently, we developed several reordering methods (spectral bisection, neighbor join tree etc.) to combine node layout and application communication data for optimized task placement. We developed a tool called mpiAproxy to facilitate detailed evaluation of the various reordering algorithms without requiring full application executions. This work presents a comprehensive performance evaluation (14,000 experiments) of the various task mapping techniques in lowering communication costs on Titan, the leadership class supercomputer at Oak Ridge National Laboratory.

97 MATHEMATICS AND COMPUTING↗

A two-level GPU-accelerated incomplete LU preconditioner for general sparse linear systems

This paper presents a parallel preconditioning approach based on incomplete LU (ILU) factorizations in the framework of Domain Decomposition (DD) for general sparse linear systems. We focus on distributed memory parallel architectures, specifically, those that are equipped with graphic processing units (GPUs). In addition to block-Jacobi, we present general purpose two-level ILU Schur complement-based approaches, where different strategies are presented to solve the coarse-level reduced system. These strategies are combined with modified ILU methods in the construction of the coarse-level operator, in order to effectively remove smooth errors by targeting an algebraically smooth vector. We leverage available GPU-based sparse matrix kernels to accelerate the setup and the solve phases of the proposed ILU preconditioner. We evaluate the efficiency of the proposed methods as a smoother for algebraic multigrid (AMG) and as a preconditioner for Krylov subspace methods on challenging anisotropic diffusion problems and a collection of general sparse matrices.

97 MATHEMATICS AND COMPUTING↗

Developing highly scalable fluid solvers for enabling multiphysics simulation.

We performed an investigation into explicit algorithms for the simulation of incompressible flows using methods with a finite, but small amount of compressibility added. Such methods include the artificial compressibility method and the lattice-Boltzmann method. The impetus for investigating such techniques stems from the increasing use of parallel computation at all levels (processors, clusters, and graphics processing units). Explicit algorithms have the potential to leverage these resources. In our investigation, a new form of artificial compressibility was derived. This method, referred to as the Entropically Damped Artificial Compressibility (EDAC) method, demonstrated superior results to traditional artificial compressibility methods by damping the numerical acoustic waves associated with these methods. Performance nearing that of the lattice- Boltzmann technique was observed, without the requirement of recasting the problem in terms of particle distribution functions; continuum variables may be used. Several example problems were investigated using a finite-di erence and finite-element discretizations of the EDAC equations. Example problems included lid-driven cavity flow, a convecting Taylor-Green vortex, a doubly periodic shear layer, freely decaying turbulence, and flow over a square cylinder. Additionally, a scalability study was performed using in excess of one million processing cores. Explicit methods were found to have desirable scaling properties; however, some robustness and general applicability issues remained.

Clausen, Jonathan R↗

Simulating of magnetic reconnection in solar flares

Magnetic reconnection is an astrophysical process where neighboring magnetic field lines running anti-parallel are reconfigured. Observations of solar flares are what pushed for serious research regarding magnetic reconnection as it seemed to be the underlying mechanism. The reconfiguration of the field lines results in an explosive release of energy as magnetic field energy is converted into plasma kinetic and thermal energies. Our work begins with Athena++, an astrophysical magnetohydrodynamic (MHD) simulation code, and an existing magnetic reconnection setup.

79 ASTRONOMY AND ASTROPHYSICS↗

Simulating ‘Two Ribbon’ Type Solar Flares [Slides]

The underlying mechanism in solar flares in a process known as magnetic reconnection. This astrophysical process is when magnetic field lines, running anti-parallel, break and reconnect. The change in configuration of field lines results in an explosive release of energy as the leftover magnetic energy is converted to kinetic and thermal energies. In this project, we are using an astrophysical magnetohydrodynamic (MHD) simulation code known as Athena++ with a reconnection problem generator file created by Li et al. 2018. Our current research is to successfully implement a radiative cooling term into the MHD equations as it could have important physical effects on the plasma. We use Klimchuk et al. 2008 and SPEX_DM as our chosen cooling functions. The results from the SPEX_DM function show there is a condensation forming that we had predicted but had not seen before with our simulation code. We will continue to analyze the SPEX_DM function and potentially implement thermal conduction.

79 ASTRONOMY AND ASTROPHYSICS↗

Simulating Magnetic Reconnection in ‘Two Ribbon’ Type Solar Flares

Magnetic reconnection is an astrophysical process where neighboring magnetic field lines, facing anti-parallel, are reconfigured. This reconfiguration results in built up magnetic energy being explosively released as it is being converted to plasma kinetic and thermal energies. There are various kinds of simulations used to simulation reconnection; our work begins with Athena++, a magnetohydrodynamic (MHD) simulation code typically used for astrophysical problems, and a reconnection specific code file. The resistive MHD equations are solved with Riemann solvers. There was an initial test run without modifying the code to understand the dynamics of the simulation. We expand on the original reconnection problem file by implementing a radiative cooling term specific to the corona. The radiative cooling is theorized to have an effect on solar coronal plasma and magnetic reconnection dynamics. The cooling term will be tested with various parameters and compared to the case without cooling to study these dynamics. The condensation found in only the with cooling case emphasizes the importance of implementing this feature and will be later tested with a thermal conduction term. We want to determine the parameter regime where non-equilibrium cooling will be important for the reconnection dynamics.

79 ASTRONOMY AND ASTROPHYSICS↗

Applying Time-Parallelization to Turbulent Flows

Parallelization of the temporal domain is explored for the solution of turbulent flows. Multigrid reduction-in-time (MGRIT) is used to advance the large-scale fluid dynamics in time sequentially on the coarsest space-time grid but propagate the information in time parallel on all other levels. The goal of this process is to accurately and efficiently resolve the coarse-scale turbulence structure and use that to drive the fine-scales of the turbulent flow. The extra forcing from nonlinear multigrid facilitates the coupling and interaction between fine and coarse scales, through which the multiscale nonlinear physics is properly captured. Adaptive mesh refinement is employed to finely resolve only the regions with strong gradients, which provides further computational efficiency. The underlying computational fluid dynamics solver is a fourth-order finite-volume scheme with the standard 4-stage Runge-Kutta method. An advanced approach is devised and implemented to enable MGRIT to solve highly turbulent flows successfully. Furthermore, the method is applied to solve a Taylor-Green vortex problem and a doubleshear-layer turbulent mixing flow. Results are promising, validating that MGRIT with the filtering approach has the potential to efficiently solve general turbulent flows.

Computational Fluid Dynamics↗

Porting HEP Parameterized Calorimeter Simulation Code to GPUs

The High Energy Physics (HEP) experiments, such as those at the Large Hadron Collider (LHC), traditionally consume large amounts of CPU cycles for detector simulations and data analysis, but rarely use compute accelerators such as GPUs. As the LHC is upgraded to allow for higher luminosity, resulting in much higher data rates, purely relying on CPUs may not provide enough computing power to support the simulation and data analysis needs. As a proof of concept, we investigate the feasibility of porting a HEP parameterized calorimeter simulation code to GPUs. We have chosen to use FastCaloSim, the ATLAS fast parametrized calorimeter simulation. While FastCaloSim is sufficiently fast such that it does not impose a bottleneck in detector simulations overall, significant speed-ups in the processing of large samples can be achieved from GPU parallelization at both the particle (intra-event) and event levels; this is especially beneficial in conditions expected at the high-luminosity LHC, where extremely high per-event particle multiplicities will result from the many simultaneous proton-proton collisions. We report our experience with porting FastCaloSim to NVIDIA GPUs using CUDA. A preliminary Kokkos implementation of FastCaloSim for portability to other parallel architectures is also described.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Characterization of submicron particles influenced by mixed biogenic and anthropogenic emissions using high-resolution aerosol mass spectrometry: results from CARES

The Carbonaceous Aerosols and Radiative Effects Study (CARES) took place in the Sacramento Valley of California in summer 2010. We present results obtained at Cool, CA, the T1 site of the project ({approx}40 km downwind of urban emissions from Sacramento), where we deployed an Aerodyne high resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS) in parallel with complementary instrumentation to characterize the sources and processes of submicron particles (PM1). Cool is located at the foothill of the Sierra Nevada Mountains, where intense biogenic emissions are periodically mixed with urban outflow transported by daytime southwesterly winds from the Sacramento metropolitan area. The particle mass loading was low (3.0 {micro}gm{sup -3} on average) and dominated by organics (80% of the PM1 mass) followed by sulfate (9.9 %). Organics and sulfate appeared to be externally mixed, as suggested by their different time series (r2 = 0.13) and size distributions. Sulfate showed a bimodal distribution with a droplet mode peaking at {approx}400nm in vacuum aerodynamic diameter (Dva), and a condensation mode at {approx}150 nm, while organics generally displayed a broad distribution in 60-600nm (Dva). New particle formation and growth events were observed almost every day, emphasizing the roles of organics and sulfate in new particle growth, especially that of organics. The organic aerosol (OA) had a nominal formula of C{sub 1}H{sub 1.38}N{sub 0.004}O{sub 0.44}, thus an average organic mass-to-carbon (OM/OC) ratio of 1.70. Two different oxygenated OA (OOA, 90% of total OA mass) and a hydrocarbon-like OA (HOA, 10 %) were identified by Positive matrix factorization (PMF) of the high resolution mass spectra. The more oxidized MO-OOA (O/C = 0.54) corresponded to secondary OA (SOA) primarily influenced by biogenic emissions, while the less oxidized LO-OOA (O/C = 0.42) corresponded to SOA associated with urban transport. The HOA factor corresponded to primary emissions mainly due to local traffic. Twenty three periods of urban plumes from T0 (Sacramento) to T1 (Cool) were identified using the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem). The average PM1 mass loading was much higher in urban plumes (3.9 {micro}gm{sup -3}) than in air masses dominated by biogenic SOA (1.8 {micro}gm{sup -3}). The change in OA mass relative to CO ({Delta}OA/{Delta}CO) varied in the range of 5-196 {micro}gm{sup -3} ppm{sup -1}, reflecting large variability in SOA production. The highest {Delta}OA/{Delta}CO were reached when urban plumes arrived at Cool in the presence of a high concentration of biogenic volatile organic compounds (BVOCs=isoprene+monoterpenes+2-methyl-3-buten-2- ol [MBO]+methyl chavicol). This ratio, which was 77 {micro}gm{sup -3} ppm{sup -1} on average when BVOCs > 2 ppb, is much higher than when urban plumes arrived in a low biogenic VOCs environment (28 {micro}gm{sup -3} ppm{sup -1} when BVOCs < 0.7 ppb) or during other periods dominated by biogenic SOA (40 {micro}gm{sup -3} ppm{sup -1}). The results from this study demon10 strate that SOA formation is enhanced when anthropogenic emissions interact with biogenic precursors.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Intranode data communications in a parallel computer

Intranode data communications in a parallel computer that includes compute nodes configured to execute processes, where the data communications include: allocating, upon initialization of a first process of a compute node, a region of shared memory; establishing, by the first process, a predefined number of message buffers, each message buffer associated with a process to be initialized on the compute node; sending, to a second process on the same compute node, a data communications message without determining whether the second process has been initialized, including storing the data communications message in the message buffer of the second process; and upon initialization of the second process: retrieving, by the second process, a pointer to the second process&#x27;s message buffer; and retrieving, by the second process from the second process&#x27;s message buffer in dependence upon the pointer, the data communications message sent by the first process.

97 MATHEMATICS AND COMPUTING↗

Multidimensional bioseparation with modular microfluidics

A multidimensional chemical separation and analysis system is described including a prototyping platform and modular microfluidic components capable of rapid and convenient assembly, alteration and disassembly of numerous candidate separation systems. Partial or total computer control of the separation system is possible. Single or multiple alternative processing trains can be tested, optimized and/or run in parallel. Examples related to the separation and analysis of human bodily fluids are given.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Intranode data communications in a parallel computer

Intranode data communications in a parallel computer that includes compute nodes configured to execute processes, where the data communications include: allocating, upon initialization of a first process of a computer node, a region of shared memory; establishing, by the first process, a predefined number of message buffers, each message buffer associated with a process to be initialized on the compute node; sending, to a second process on the same compute node, a data communications message without determining whether the second process has been initialized, including storing the data communications message in the message buffer of the second process; and upon initialization of the second process: retrieving, by the second process, a pointer to the second process's message buffer; and retrieving, by the second process from the second process's message buffer in dependence upon the pointer, the data communications message sent by the first process.

97 MATHEMATICS AND COMPUTING↗

Communication Characterization and Optimization of Applications Using Topology-Aware Task Mapping on Large Supercomputers

On large supercomputers, the job scheduling systems may assign a non-contiguous node allocation for user applications depending on available resources. With parallel applications using MPI (Message Passing Interface), the default process ordering does not take into account the actual physical node layout available to the application. This contributes to non-locality in terms of physical network topology and impacts communication performance of the application. In order to mitigate such performance penalties, this work describes techniques to identify suitable task mapping that takes the layout of the allocated nodes as well as the application's communication behavior into account. During the first phase of this research, we instrumented and collected performance data to characterize communication behavior of critical US DOE (United States - Department of Energy) applications using an augmented version of the mpiP tool. Subsequently, we developed several reordering methods (spectral bisection, neighbor join tree etc.) to combine node layout and application communication data for optimized task placement. We developed a tool called mpiAproxy to facilitate detailed evaluation of the various reordering algorithms without requiring full application executions. This work presents a comprehensive performance evaluation (14,000 experiments) of the various task mapping techniques in lowering communication costs on Titan, the leadership class supercomputer at Oak Ridge National Laboratory.

Sreepathi, Sarat↗

Two-Stage Gauss-Seidel Preconditioners and Smoothers for Krylov Solvers on a GPU Cluster: Preprint

Gauss-Seidel (GS) relaxation is often employed as a preconditioner for a Krylov solver or as a smoother for Algebraic Multigrid (AMG). However, the requisite sparse triangular solve is difficult to parallelize on many-core architectures such as graphics processing units (GPUs). In the present study, the performance of the sequential GS relaxation based on a triangular solve is compared with two-stage variants, replacing the direct triangular solve with a fixed number of inner Jacobi-Richardson (JR) iterations. When a small number of inner iterations is sufficient to maintain the Krylov convergence rate, the two-stage GS (GS2) often outperforms the sequential algorithm on many-core architectures. The GS2 algorithm is also compared with JR. When they perform the same number of ops for SpMV (e.g. three JR sweeps compared to two GS sweeps with one inner JR sweep), the GS2 iterations, and the Krylov solver preconditioned with GS2, may converge faster than the JR iterations. Moreover, for some problems (e.g. elasticity), it was found that JR may diverge with a damping factor of one, whereas two-stage GS may improve the convergence with more inner iterations. Finally, to study the performance of the two-stage smoother and preconditioner for a practical problem, these were applied to incompressible uid ow simulations on GPUs.

algebraic multigrid↗

Reducing Parallel Communication in Algebraic Multigrid through Sparsification

Algebraic multigrid (AMG) is an $\mathcal{O}(n)$ solution process for many large sparse linear systems. A hierarchy of progressively coarser grids which utilize complementary relaxation and interpolation operators is constructed. High-energy error is reduced by relaxation, while low-energy error is mapped to coarse-grid matrices and reduced there. However, large parallel communication costs often limit parallel scalability. Furthermore, as the multigrid hierarchy is formed, each coarse matrix is formed through a triple matrix product. The resulting coarse grids often have significantly more nonzeros per row than the original fine-grid operator, thereby generating high parallel communication costs associated with sparse matrix-vector multiplication (SpMV) on coarse levels. In this paper, we introduce a method that systematically removes entries in coarse-grid matrices after the hierarchy is formed, leading to improved communication costs. We sparsify by removing weakly connected or unimportant entries in the matrix, leading to improved solve time. The main trade-off is that if the heuristic identifying unimportant entries is used too aggressively, then AMG convergence can suffer. To counteract this, the original hierarchy is retained, allowing entries to be reintroduced into the solver hierarchy if convergence is too slow. This enables a balance between communication cost and convergence, as necessary. In this paper we present new algorithms for reducing communication and present a number of computational experiments in support.

97 MATHEMATICS AND COMPUTING↗

Performance modeling of microsecond scale biological molecular dynamics simulations on heterogeneous architectures

Performance improvements in biomolecular simulations based on molecular dynamics (MD) codes are widely desired. Unfortunately, the factors, which allowed past performance improvements, particularly the microprocessor clock frequencies, are no longer increasing. Hence, novel software and hardware solutions are being explored for accelerating performance of widely used MD codes. In this paper, we describe our efforts on porting, optimizing and tuning of Large-scale Atomic/Molecular Massively Parallel Simulator, a popular MD framework, on heterogeneous architectures: multi-core processors with graphical processing unit (GPU) accelerators. Our implementation is based on accelerating the most computationally expensive non-bonded interaction terms on the GPUs and overlapping the computation on the CPU and GPUs. This functionality is built on top of message passing interface that allows multi-level parallelism to be extracted even at the workstation level with the multi-core CPUs and allows extension of the implementation on GPU-enabled clusters. We hypothesize that the optimal benefit of heterogeneous architectures for applications will come by utilizing all possible resources (for example, CPU-cores and GPU devices on GPU-enabled clusters). Benchmarks for a range of biomolecular system sizes are provided, and an analysis is performed on four generations of NVIDIA's GPU devices. On GPU-enabled Linux clusters, by overlapping and pipelining computation and communication, we observe up to 10-folds application acceleration in multi-core and multi-GPU environments illustrating significant performance improvements. Detailed analysis of the implementation is presented that allows identification of bottlenecks in algorithm, indicating that code optimization and improvements on GPUs could allow microsecond scale simulation throughput on workstations and inexpensive GPU clusters, putting widely desired biologically relevant simulation time-scales within reach of a large user community. In order to systematically optimize simulation throughput and to enable performance prediction, we have developed a parameterized performance model that will allow developers and users to explore the performance potential of future heterogeneous systems for biological simulations.

Agarwal, Pratul K.↗

Pseudo-dynamic Aspen Plus{sup R} Model of the Defense Waste Processing Facility - 18475

The Defense Waste Processing Facility (DWPF) at the Savannah River Site converts legacy high-level waste from the Cold War into borosilicate glass for permanent disposal in a geological repository. The waste sludge is chemically adjusted by the addition of acid, mixed with frit, and concentrated in the DWPF Chemical Process Cell (CPC) to ensure that the final glass has the desired properties. The resulting mixture is fed into the melter where the waste is melted and poured into stainless steel storage canisters. We present an Aspen Plus{sup R} model of the DWPF CPC, which is being developed to aid chemical processing decisions pertaining to the current flowsheet and to aid decisions regarding optimizing future processing. Given a waste feed composition with acid and frit additions, the model will predict the sludge and off-gas compositions throughout the processing, as well as the final melter feed composition. This allows different processing scenarios to be investigated easily, reducing the reliance on experimental testing, and hence saving time and money. The model has multiple potential applications, including calculating the acid and frit additions required to give acceptable glass properties, off-gas permitting, tracking mercury steam-stripped from the waste in the off-gas system, tracking flammable gases in the offgas system, investigating the effect of reducing air purges in the condensers, and general process optimization. The waste is processed in the CPC in batches, requiring dynamic simulation to track the time-dependent operations. Accurate modeling of the processing also requires calculations of phase and electrolyte chemical equilibria and parallel rate-dependent reactions. Due to existing technical limitations with dynamic simulation of the complex chemical system, we have adopted a pseudo-dynamic approach in which time intervals are treated as consecutive steady-state operations in Aspen Plus{sup R}. Thus, the model consists of ∼100 replicas of the processing vessel and its off-gas system, with the sludge output by one replica (representing one time interval) becoming the feed to the next (representing the next time interval). By examining the results from these replicas in sequence, one can see the time evolution of the model. This general approach can be used wherever simultaneous calculation of phase and electrolyte chemical equilibria in a dynamically evolving system is needed. In addition to describing the general structure and chemistry of the model, we present results from tests of the model against experimental data. These data are from laboratory-scale process simulations of the CPC with a nonradioactive waste sludge simulant. These tests include the model's predictions of the offgas composition, the sludge pH, and the final sludge composition. We will also discuss planned future developments of the model. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

PARTISN on Advanced/Heterogeneous Processing Systems [Slides]

This presentation describes both the importance and the difficulty of parallelizing deterministic neutron transport calculations using the LANL code PARTISN. A brief history of the standard "KBA" parallelization method used for deterministic SN transport is presented, then the modifications necessary to implement this on Roadrunner are covered. The use of the Roadrunner algorithm with OpenMP is then described, along with a brief mention of an alternative thermal radiation transport algorithm for GPUs. Finally, ongoing work with PARTISN on the MIC many-core architecture is noted.

97 MATHEMATICS AND COMPUTING↗