Search NASA⌕ Search

SEARCH · Search NASA

Results for “Parallel Architectures”

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 1,765 records · Page 98
All sourcesNASA NTRSDOE OSTI
NASA NTRS2014

Refueling with In-Situ Produced Propellants

In-situ produced propellants have been identified in many architecture studies as key to implementing feasible chemical propulsion missions to destinations beyond lunar orbit. Some of the more noteworthy ones include: launching from Mars to return to Earth (either direct from the surface, or via an orbital rendezvous); using the Earth-Moon Lagrange point as a place to refuel Mars transfer stages with Lunar surface produced propellants; and using Mars Moon Phobos as a place to produce propellants for descent and ascent stages bound for the Mars surface. However successful implementation of these strategies require an ability to successfully transfer propellants from the in-situ production equipment into the propellant tankage of the rocket stage used to move to the desired location. In many circumstances the most desirable location for this transfer to occur is in the low-gravity environment of space. In support of low earth orbit propellant depot concepts, extensive studies have been conducted on transferring propellants in-space. Most of these propellant transfer techniques will be applicable to low gravity operations in other locations. Even ground-based transfer operations on the Moon, Mars, and especially Phobos could benefit from the propellant conserving techniques used for depot refueling. This paper will review the literature of in-situ propellants and refueling to: assess the performance benefits of the use in-situ propellants for mission concepts; review the parallels with propellant depot efforts; assess the progress of the techniques required; and provide recommendations for future research.

In-Situ Propellants↗
DOE OSTI2020

QCOR; A Language Extension Specification for the Heterogeneous Quantum-Classical Model of Computation

Quantum computing (QC) is an emerging computational paradigm that leverages the laws of quantum mechanics to perform elementary logic operations. Existing programming models for QC were designed with fault-tolerant hardware in mind, envisioning stand-alone applications. However, the susceptibility of near-term quantum computers to noise limits their stand-alone utility. To better leverage limited computational strengths of noisy quantum devices, hybrid algorithms have been suggested whereby quantum computers are used in tandem with their classical counterparts in a heterogeneous fashion. This modus operandi calls out for a programming model and a high-level programming language that natively and seamlessly supports heterogeneous quantum-classical hardware architectures in a single-source-code paradigm. Motivated by the lack of such a model, we introduce a language extension specification, called QCOR, which enables single-source quantum-classical programming. Programs written using the QCOR library–based language extensions can be compiled to produce functional hybrid binary executables. After defining QCOR’s programming model, memory model, and execution model, we discuss how QCOR enables variational, iterative, and feed-forward QC. Additionally, QCOR approaches quantum-classical computation in a hardware-agnostic heterogeneous fashion and strives to build on best practices of high-performance computing. The high level of abstraction in the language extension is intended to accelerate the adoption of QC by researchers familiar with classical high-performance computing.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Optimizing the hypre solver for manycore and GPU architectures

The solution of large-scale combustion problems with codes such as Uintah on modern computer architectures requires the use of multithreading and GPUs to achieve performance. Uintah uses a low-Mach number approximation that requires iteratively solving a large system of linear equations. The Hypre iterative solver has solved such systems in a scalable way for Uintah, but the use of OpenMP with Hypre leads to at least slowdown due to OpenMP overheads. The proposed solution uses the MPI Endpoints within Hypre, where each team of threads acts as a different MPI rank. This approach minimizes OpenMP synchronization overhead and performs as fast or (up to 1.44) faster than Hypre's MPI-only version, and allows the rest of Uintah to be optimized using OpenMP. The profiling of the GPU version of Hypre shows the bottleneck to be the launch overhead of thousands of micro-kernels. The GPU performance was improved by fusing these micro-kernels and was further optimized by using Cuda-aware MPI, resulting in an overall speedup of 1.16—1.44 compared to the baseline GPU implementation. The above optimization strategies were published in the International Conference on Computational Science 2020 [1]. This work extends the previously published research by carrying out the second phase of communication-centered optimizations in Hypre to improve its scalability on large-scale supercomputers. Additionally, this includes an efficient non-blocking inter-thread communication scheme, communication-reducing patch assignment, and expression of logical communication parallelism to a new version of the MPICH library that utilizes the underlying network parallelism [2]. The above optimizations avoid communication bottlenecks previously observed during strong scaling and improve performance by up to 2 on 256 nodes of Intel Knight's Landing processor.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

How the Size and Density of Charge-Transfer Excitons Depend on Heterojunction’s Architecture

We have characterized the size, intensity, density, and distribution of charge-transfer (CT) excitons as a function of the acceptor–donor architecture of prototypical organic interfaces. This characterization was done by computational analysis of 17 models of varying numbers, positions, and orientations of the donor and acceptor molecules. The models’ building blocks were phenyl-C61-butyric acid methyl ester (PCBM) fullerene acceptors and dual-band donor polymers composed of thiophene, benzothiadiazole, and benzotriazole subunits. The electronic structure of the donor–acceptor complexes was computed with the time-dependent long-range-corrected density-functional tight-binding method and analyzed with the fragment-based one-electron transition density matrix. In all models, the complexes with edge-on orientation have denser spectra of low-energy CT states lying below the absorption bands compared to the complexes with face-on orientation. This CT-state distribution in edge-on complexes provides a gate to efficiently populate cold CT excitons. Moreover, the cold CT excitons have a higher degree of charge separation in the edge-on than in the face-on complexes. The CT amount and the CT exciton size generally increase with the energy of the CT states, although the electron remains localized on a single molecule in cold CT states. Delocalization over two PCBM molecules was observed for high-energy CT states. The exciton size also depends on the orientation. Larger excitons are produced by the delocalization of the electrons perpendicularly to the interface. When the delocalization is parallel, the smaller electron–hole distances yield moderately sized CT excitons.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

Scalable training of trustworthy and energy-efficient predictive graph foundation models for atomistic materials modeling: a case study with HydraGNN

We present our work on developing and training scalable, trustworthy, and energy-efficient predictive graph foundation models (GFMs) using HydraGNN, a multi-headed graph convolutional neural network architecture. HydraGNN expands the boundaries of graph neural network (GNN) computations in both training scale and data diversity. It abstracts over message passing algorithms, allowing both reproduction of and comparison across algorithmic innovations that define nearest-neighbor convolution in GNNs. This work discusses a series of optimizations that have allowed scaling up the GFMs training to tens of thousands of GPUs on datasets consisting of hundreds of millions of graphs. Our GFMs use multitask learning (MTL) to simultaneously learn graph-level and node-level properties of atomistic structures, such as energy and atomic forces. Using over 154 million atomistic structures for training, we illustrate the performance of our approach along with the lessons learned on two state-of-the-art US Department of Energy (US-DOE) supercomputers, namely the Perlmutter petascale system at the National Energy Research Scientific Computing Center and the Frontier exascale system at Oak Ridge Leadership Computing Facility. The HydraGNN architecture enables the GFM to achieve near-linear strong scaling performance using more than 2000 GPUs on Perlmutter and 16,000 GPUs on Frontier.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Residential Building Energy Efficiency Field Studies: Low-Rise Multifamily

In recent years, the U.S. Department of Energy (DOE) has conducted a series of research studies to validate energy efficient building technologies in the field. Much of the work has focused on single-family construction, and some has also addressed commercial energy codes. The work detailed in this DOE-funded study (EE0007616) focuses on low-rise multifamily buildings (three stories or fewer above grade) in various regions of the United States, and reports on how state-level building codes are being implemented, both in terms of observed characteristics and also in terms of estimated energy impacts. Nearly 100 buildings across four states—Illinois, Minnesota, Oregon, and Washington—were sampled, which represent a range of climate types from mild temperature to very cold continental. Both common entry and outdoor entry buildings were included, and a parallel research project evaluated envelope air tightness and current still-evolving air tightness testing methods. Finally, a set of structured interviews of building designers and other relevant professionals was carried to out to gain more insight into this market. To the greatest extent possible, the methodology developed under the project for low-rise multifamily buildings mirrored the approach established by Pacific Northwest National Laboratory (PNNL) for single-family residential buildings (https://www.energy.gov/eere/buildings/downloads/residential-building-energy-code-field-study). This included the general approach to sampling, recruitment, and data collection, as well as data analysis and presentation. The range of permitting dates for the sites encompassed two energy code cycles in most regions. All states in the study had adopted a variation of the International Energy Conservation Code (IECC) for the structure of their state code. The low-rise multifamily occupancy presents a hybrid building type: most of the building’s conditioned floor area was covered by the residential chapter of the code while portions of the building (such as corridors and common spaces) fell under the commercial code chapter. The key items assessed in this work were: Building Shell—exterior wall insulation, ceiling insulation, foundation insulation, windows. Common Areas—HVAC and lighting. Living Units—lighting, ventilation. A few items were not assessed in detail, given their relative paucity in this occupancy type; these included duct leakage, pipe insulation, and hot water circulation controls. Building characteristics were collected via a combination of architectural, mechanical, electrical, and plumbing plan reviews and field inspections, and entered into a spreadsheet-based tool that was later queried to build a database. Data went through quality control both upon arrival and via a later semi-automated review and assurance process. Most of the data are presented graphically so that the reader can quickly assess compliance with the applicable energy codes (both by state and by code year). As a final step, EnergyPlus™ simulations were created for all buildings in the study to estimate both the as-found energy use intensity (EUI) and the energy and CO 2 that could be saved if features that were found to not meet code minimums were brought up to code. The savings estimates were tabulated for each of the four states in the study. The research team found that the single-family approach was largely applicable to low-rise multifamily buildings. This applies to both the data collection and the prototype EUI analysis. Most of the occupied space is living units and falls under residential energy codes, and many characteristics use similar envelope construction and relatively straightforward mechanical systems and lighting. One of the most challenging aspects of this work was to build an effective spreadsheet-based data collection instrument that could allow efficient collection of both building plan and field data. The research team is of the view that other methods could be equally effective if the work is done carefully with diligent quality control. The primary findings for the work center around the thermal envelope and mechanical systems and lighting at the sites: For thermal envelope components, the majority of buildings met or were better than the prescriptive code.This suggests that building designers and builders are aware of code requirements. In some cases, surveyed buildings were designed to qualify for energy efficiency certification programs. These buildings made up at least 20% of sampled buildings in each state. Almost all buildings met mechanical system efficiency requirements (for both living units and common areas). In some cases, sites employed systems that were considerably more efficient than required by the applicable energy code. Dwelling units had a majority of high-efficacy lighting, often in excess of the state’s residential code requirements. While high-efficacy fixtures were also typical in common areas (corridors and stairwells), lighting power densities (LPDs) in these areas were sometimes higher than levels dictated by the applicable part of the state commercial energy code. The simulation models run on a series of low-rise multifamily prototypes, informed by a composite of the field data collected, calculated annual EUIs of between 20 and 50 kBtu/ft2-yr, with the range representing the effects of both building characteristics and building location (climate zone). A detailed process (based on simulations of prototype buildings) was used to estimate the amount of avoided energy use that would occur if 100% adherence to energy codes were attained. The results indicated modest savings are attainable for items such as window thermal performance and common area lighting. The result is overall only a modest potential for additional energy savings, averaging about 10% of EUI.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
NASA NTRS2013

The Antarctic Search for Meteorites: A Model for Deep Space Exploration

In an era of shrinking buying power and reduced flight opportunities, NASA must extract the greatest possible value from all sources of insight into the future of human space exploration. Antarctica is one such source. The history of Antarctic exploration has many political and technical parallels with the development of space, and Antarctica's remoteness and harsh climate make it an excellent proxy for space (e.g., [1,2]). Links between exploration of space and of the Antarctic date back to the International Geophysical Year of 1957-1958, which saw both the launch of Sputnik 1 and the establishment of a station at the South Pole. The Antarctic Search for Meteorites (ANSMET) is an annual expedition to the south polar plateau to collect meteorites. Although its intent is not to simulate a space mission, the handful of astronauts who have participated in ANSMET agree that it is very similar to a long-duration space flight. Independently, NASA and other space agencies have simulated deep space exploration missions in "analog" activities at remote field sites on Earth (e.g., [3]). These include NASA Extreme Environment Mission Operations (NEEMO) [4,5], Desert Research And Technology Studies (Desert RATS) [6,7], and the Pavilion Lake Research Project (PLRP or simply Pavilion Lake) [8]. This report focuses on NEEMO, Desert RATS, and PLRP because of the author's firsthand experience with them. Other noteworthy analogs, such as the arctic Haughton Mars Project and the European Space Agency's underground Cooperative Adventure for Valuing and Exercising human behavior and performance Skills (CAVES), are not treated here. NASA analogs often include fully staffed control centers, astronauts serving as crew, and realistic mission timelines lasting one to two weeks. Analogs have provided key insights into system architectures and operational concepts for the future human exploration of deep space. They have pioneered techniques for human communication with significant speed-of-light delays, for conducting spacewalks on natural objects with negligible surface gravity, and for empowering exploration crews to work with reduced dependence on a ground control center. They have field-tested dozens of emerging technologies including spacewalking tools and full-scale prototype vehicles and habitats. They have provided valuable experience for astronauts preparing for their first space flight, and for flown crewmembers who will take command roles on later flights. Some analogs, especially PLRP, have connected observers in the field with science teams in remotely located control centers to produce high-quality, publishable scientific results. The analogs have accomplished all of this at a tiny fraction of the cost of an actual space flight. This report treats ANSMET as space flight analog. The chapter following this introduction describes ANSMET in depth. The report then presents data on logistics and crew considerations that may be useful for developers of future human space exploration missions. It offers detailed comparisons between ANSMET and past, present, and future space flights on the Space Shuttle, the International Space Station (ISS), and a proposed Mars mission. Those comparisons are intended to complement the work of Eppler [2], who compares ANSMET to the Apollo moon flights. This report also compares ANSMET with the Desert RATS, NEEMO, and PLRP analogs. It then presents observations and makes recommendations related to ANSMET's value as a simulated space mission. The report ends with a short conclusion. The remainder of this introductory chapter provides background material to help readers interpret the rest of the report. It gives brief overviews of Space Shuttle and ISS missions along with information on a notional future human flight to Mars. It also presents the general features of three of NASA's space flight "analog" projects. With those points of reference in place, the chapter concludes with an overview of ANSMET.

Love, Stanley G.↗
DOE OSTI2020

Fast GPU 3D diffeomorphic image registration

3D image registration is one of the most fundamental and computationally expensive operations in medical image analysis. Here, we present a mixed-precision, Gauss–Newton–Krylov solver for diffeomorphic registration of two images. Our work extends the publicly available CLAIRE library to GPU architectures. Despite the importance of image registration, only a few implementations of large deformation diffeomorphic registration packages support GPUs. Our contributions are new algorithms to significantly reduce the run time of the two main computational kernels in CLAIRE: calculation of derivatives and scattered-data interpolation. Additionally, we deploy (i) highly-optimized, mixed-precision GPU-kernels for the evaluation of scattered-data interpolation, (ii) replace Fast-Fourier-Transform (FFT)-based first-order derivatives with optimized 8th-order finite differences, and (iii) compare with state-of-the-art CPU and GPU implementations. As a highlight, we demonstrate that we can register clinical images in less than 6 s on a single NVIDIA Tesla V100. This amounts to over 20 speed-up over the current version of CLAIRE and over 30 speed-up over existing GPU implementations.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Locality-Aware Scheduling for Scalable Heterogeneous Environments

Heterogeneous computing promise boost performance of scientific applications by allowing massively parallel execution of computational tasks. However, manually managing extremely heterogeneous, multi-device systems is complicated and may result in sub-optimal performance. Specifically, data management is an extremely challenging problem on multi-device systems. In this work, we introduce two locality-aware schedulers for the Minos Computing Library (MCL), an asynchronous, task-based programming model and runtime for extremely heterogeneous systems. The first scheduler implements a pure locality-aware algorithm to maximize data reuse, though it might incur in ”hot-spots” that limit system utilization. The second scheduler mitigates this drawback by dynamically targeting between locality-awareness and system utilization based on the current workload and available computing devices. Our results show that locality-awareness greatly benefit applications that exhibit data reuse, providing up to 6.9x and 7.9x over the original MCL scheduler and equivalent OpenCL implementations, respectively. Moreover, our schedulers introduce negligible overhead compared with the original MCL scheduler and achieve similar performance for applications that don’t benefit from data locality.

Architecture, co-design, Task-based programming mo↗
NASA NTRS1990

Macromolecular differentiation of Golgi stacks in root tips of Arabidopsis and Nicotiana seedlings as visualized in high pressure frozen and freeze-substituted samples

The plant root tip represents a fascinating model system for studying changes in Golgi stack architecture associated with the developmental progression of meristematic cells to gravity sensing columella cells, and finally to "young" and "old", polysaccharide-slime secreting peripheral cells. To this end we have used high pressure freezing in conjunction with freeze-substitution techniques to follow developmental changes in the macromolecular organization of Golgi stacks in root tips of Arabidopsis and Nicotiana. Due to the much improved structural preservation of all cells under investigation, our electron micrographs reveal both several novel structural features common to all Golgi stacks, as well as characteristic differences in morphology between Golgi stacks of different cell types. Common to all Golgi stacks are clear and discrete differences in staining patterns and width of cis, medial and trans cisternae. Cis cisternae have the widest lumina (approximately 30 nm) and are the least stained. Medial cisternae are narrower (approximately 20 nm) and filled with more darkly staining products. Most trans cisternae possess a completely collapsed lumen in their central domain, giving rise to a 4-6 nm wide dark line in cross-sectional views. Numerous vesicles associated with the cisternal margins carry a non-clathrin type of coat. A trans Golgi network with clathrin coated vesicles is associated with all Golgi stacks except those of old peripheral cells. It is easily distinguished from trans cisternae by its blebbing morphology and staining pattern. The zone of ribosome exclusion includes both the Golgi stack and the trans Golgi network. Intercisternal elements are located exclusively between trans cisternae of columella and peripheral cells, but not meristematic cells. In older peripheral cells only trans cisternae exhibit slime-related staining. Golgi stacks possessing intercisternal elements also contain parallel rows of freeze-fracture particles in their trans cisternal membranes. We propose that intercisternal elements serve as anchors of enzyme complexes involved in the synthesis of polysaccharide slime molecules to prevent the complexes from being dragged into the forming secretory vesicles by the very large slime molecules. In addition, we draw attention to the similarities in composition and apparent site of synthesis of xyloglucans and slime molecules.

NASA Program Space Biology↗
NASA NTRS1997

Efficacy of Code Optimization on Cache-based Processors

The current common wisdom in the U.S. is that the powerful, cost-effective supercomputers of tomorrow will be based on commodity (RISC) micro-processors with cache memories. Already, most distributed systems in the world use such hardware as building blocks. This shift away from vector supercomputers and towards cache-based systems has brought about a change in programming paradigm, even when ignoring issues of parallelism. Vector machines require inner-loop independence and regular, non-pathological memory strides (usually this means: non-power-of-two strides) to allow efficient vectorization of array operations. Cache-based systems require spatial and temporal locality of data, so that data once read from main memory and stored in high-speed cache memory is used optimally before being written back to main memory. This means that the most cache-friendly array operations are those that feature zero or unit stride, so that each unit of data read from main memory (a cache line) contains information for the next iteration in the loop. Moreover, loops ought to be 'fat', meaning that as many operations as possible are performed on cache data-provided instruction caches do not overflow and enough registers are available. If unit stride is not possible, for example because of some data dependency, then care must be taken to avoid pathological strides, just ads on vector computers. For cache-based systems the issues are more complex, due to the effects of associativity and of non-unit block (cache line) size. But there is more to the story. Most modern micro-processors are superscalar, which means that they can issue several (arithmetic) instructions per clock cycle, provided that there are enough independent instructions in the loop body. This is another argument for providing fat loop bodies. With these restrictions, it appears fairly straightforward to produce code that will run efficiently on any cache-based system. It can be argued that although some of the important computational algorithms employed at NASA Ames require different programming styles on vector machines and cache-based machines, respectively, neither architecture class appeared to be favored by particular algorithms in principle. Practice tells us that the situation is more complicated. This report presents observations and some analysis of performance tuning for cache-based systems. We point out several counterintuitive results that serve as a cautionary reminder that memory accesses are not the only factors that determine performance, and that within the class of cache-based systems, significant differences exist.

VanderWijngaart, Rob F.↗
DOE OSTI2019

Production Application Performance Data Streaming for System Monitoring

In this article, we present an approach to streaming collection of application performance data. Practical application performance tuning and troubleshooting in production high-performance computing (HPC) environments requires an understanding of how applications interact with the platform, including (but not limited to) parallel programming libraries such as Message Passing Interface (MPI). Several profiling and tracing tools exist that collect heavy runtime data traces either in memory (released only at application exit) or on a file system (imposing an I/O load that may interfere with the performance being measured). Although these approaches are beneficial in development stages and post-run analysis, a systemwide and low-overhead method is required to monitor deployed applications continuously. This method must be able to collect information at both the application and system levels to yield a complete performance picture. In our approach, an application profiler collects application event counters. A sampler uses an efficient inter-process communication method to periodically extract the application counters and stream them into an infrastructure for performance data collection. We implement a tool-set based on our approach and integrate it with the Lightweight Distributed Metric Service (LDMS) system, a monitoring system used on large-scale computational platforms. LDMS provides the infrastructure to create and gather streams of performance data in a low overhead manner. We demonstrate our approach using applications implemented with MPI, as it is one of the most common standards for the development of large-scale scientific applications. We utilize our tool-set to study the impact of our approach on an open source HPC application, Nalu. Our tool-set enables us to efficiently identify patterns in the behavior of the application without source-level knowledge. We leverage LDMS to collect system-level performance data and explore the correlation between the system and application events. Finally, we demonstrate how our tool-set can help detect anomalies with a low latency. We run tests on two different architectures: a system enabled with Intel Xeon Phi and another system equipped with Intel Xeon processor. Our overhead study shows our method imposes at most 0.5% CPU usage overhead on the application in realistic deployment scenarios.

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRSDate not supplied

Utilizing Gaps and Key Performance Parameters to Inform NASA Environmental Control and Life Support and Human Health and Performance Capability Technology Decisions

Human spaceflight is a complex endeavor requiring multiple capabilities for transportation, crew health, scientific goals, and safe return to Earth. The difference between spaceflight proven capabilities and those needed for future exploration architectures is defined as a capability gap. Capability gaps are not technology specific. Each capability gap is approachable with a wide array of technologies that have unique benefits and challenges. Determining what a capability’s relevant and distinguishing key performance parameters (KPPs) are for a mission is critical. Mass, power, and volume are always constrained and important, but defining these in a way normalized by performance is challenging. Additionally, KPP definition for reliability, dormancy, and integration needs are very important and still evolving. This paper provides the approach of the Environmental Control and Life Support – Crew Health and Performance (ECLSS-CHP) System Capability Leadership Team (SCLT) has used to define gaps and KPPs in support of the NASA’s Capabilities Integration Team data call objectives. The nine ECLSS-CHP capability areas are decomposed to capabilities with ~76 gaps and supported with KPPs. Rather than defining very detailed gaps, ECLSS-CHP defines high-level gaps to be technology agnostic. Within a gap, detailed KPPs are defined to both compare technologies and measure progress within a technology over time. Ideally, KPPs are clearly defined, widely communicated both internally and externally, and provide a common nomenclature to describe the state of the art and the degree of improvement required for exploration missions. KPPs help define when the gap is closed, and the core mission objectives can be accomplished. Further technology improvements to enhance the capability, as measured by improved KPPs, must then be weighed against investments in open capability gaps that prevent NASA from achieving its exploration missions. It is uncommon that a technology maturation to improve all the relevant KPPs simultaneously but using KPPs is a critical technology investment decision making component. In addition to traditional technology selections, KPPs are informing how investments in ground testing prior to and in parallel with ISS technology demonstrations are required to improve reliability KPPs. The collection of all major technology activities within a capability area are captured on technology roadmaps to communicate how diverse program activities are coordinated to close gaps and infuse into exploration mission needs. A selection of ECLSS-CHP gaps and KPPs and their formulation, current state, and how they inform capability roadmap planning are discussed.

Life Support↗
DOE OSTI2023

Distributed out-of-memory NMF on CPU/GPU architectures

We propose an efficient distributed out-of-memory implementation of the non-negative matrix factorization (NMF) algorithm for heterogeneous high-performance-computing systems. The proposed implementation is based on prior work on NMFk, which can perform automatic model selection and extract latent variables and patterns from data. In this work, we extend NMFk by adding support for dense and sparse matrix operation on multi-node, multi-GPU systems. The resulting algorithm is optimized for out-of-memory problems where the memory required to factorize a given matrix is greater than the available GPU memory. Memory complexity is reduced by batching/tiling strategies, and sparse and dense matrix operations are significantly accelerated with GPU cores (or tensor cores when available). Input/output latency associated with batch copies between host and device is hidden using CUDA streams to overlap data transfers and compute asynchronously, and latency associated with collective communications (both intra-node and inter-node) is reduced using optimized NVIDIA Collective Communication Library (NCCL) based communicators. Benchmark results show significant improvement, from 32X to 76x speedup, with the new implementation using GPUs over the CPU-based NMFk. Good weak scaling was demonstrated on up to 4096 multi-GPU cluster nodes with approximately 25,000 GPUs when decomposing a dense 340 Terabyte-size matrix and an 11 Exabyte-size sparse matrix of density 10 -6 .

97 MATHEMATICS AND COMPUTING↗
NASA NTRS2000

Web Exploration Tools for a Fast Federated Optical Survey Database

We implemented several new web-based tools to improve the efficiency and versatility of access to the APS Catalog of the POSS I (Palomar Observatory-National Geographic Sky Survey) and its associated image database. The most important addition was a federated database system to link the APS Catalog and image database into one Internet-accessible database. With the FDBS, the queries and transactions on the integrated database are performed as if it were a single database. We installed Myriad the FDBS developed by Professor Jaideep Srivastava and members of his group in the University of Minnesota Computer Science Department. It is the first system to provide schema integration, query processing and optimization, and transaction management capabilities in a single framework. The attached figure illustrates the Myriad architecture. The FDBS permits horizontal access to the data, not just vertical. For example, for the APS, queries can be made not only by sky position, but also by any parameter present in either of the databases. APS users will be able to produce an image of all the blue galaxies and stellar sources for comparison with x-ray source error ellipses from AXAF (X Ray Astrophysics Facility) (Chandra) for example. The FDBS is now available as a beta release with the appropriate query forms at our web site. While much of our time was occupied with adapting Myriad to the APS environment, we also made major changes in Star Base, our DBMS for the Catalog, at the web interface to improve its efficiency for issuing and processing queries. Star Base is now three times faster for large queries. Improvements were also made at the web end of the image database for faster access; although work still needs to be done to the image database itself for more efficient return with the FDBS. During the past few years, we made several improvements to the database pipeline that creates the individual plate databases queries by StarBase. The changes include improved positions especially for galaxies, using a new median centroider and integrated magnitudes for galaxies with an improved density-to-intensity calibration with a "sky" background subtraction. In the original version of StarBase the object classification fainter than 19.5-20.0 mag., was an extrapolation of the networks trained on brighter objects. We have used a new catalog of galaxies at the NGP to train a neural network on objects fainter than 20th mag. This improved classification is used in the new version of StarBase. We have also added a FITS table option for the returned data from queries on the object catalog. The APS image database includes images in both colors so we have added a tool for querying the image database in both colors simultaneously. The images can be displayed in parallel or blinked for comparison.

Humphreys, Roberta M.↗
NASA NTRS2016

Investigating the Simulink Auto-Coding Process

Model based program design is the most clear and direct way to develop algorithms and programs for interfacing with hardware. While coding "by hand" results in a more tailored product, the ever-growing size and complexity of modern-day applications can cause the project work load to quickly become unreasonable for one programmer. This has generally been addressed by splitting the product into separate modules to allow multiple developers to work in parallel on the same project, however this introduces new potentials for errors in the process. The fluidity, reliability and robustness of the code relies on the abilities of the programmers to communicate their methods to one another; furthermore, multiple programmers invites multiple potentially differing coding styles into the same product, which can cause a loss of readability or even module incompatibility. Fortunately, Mathworks has implemented an auto-coding feature that allows programmers to design their algorithms through the use of models and diagrams in the graphical programming environment Simulink, allowing the designer to visually determine what the hardware is to do. From here, the auto-coding feature handles converting the project into another programming language. This type of approach allows the designer to clearly see how the software will be directing the hardware without the need to try and interpret large amounts of code. In addition, it speeds up the programming process, minimizing the amount of man-hours spent on a single project, thus reducing the chance of human error as well as project turnover time. One such project that has benefited from the auto-coding procedure is Ramses, a portion of the GNC flight software on-board Orion that has been implemented primarily in Simulink. Currently, however, auto-coding Ramses into C++ requires 5 hours of code generation time. This causes issues if the tool ever needs to be debugged, as this code generation will need to occur with each edit to any part of the program; additionally, this is lost time that could be spent testing and analyzing the code. This is one of the more prominent issues with the auto-coding process, and while much information is available with regard to optimizing Simulink designs to produce efficient and reliable C++ code, not much research has been made public on how to reduce the code generation time. It is of interest to develop some insight as to what causes code generation times to be so significant, and determine if there are architecture guidelines or a desirable auto-coding configuration set to assist in streamlining this step of the design process for particular applications. To address the issue at hand, the Simulink coder was studied at a foundational level. For each different component type made available by the software, the features, auto-code generation time, and the format of the generated code were analyzed and documented. Tools were developed and documented to expedite these studies, particularly in the area of automating sequential builds to ensure accurate data was obtained. Next, the Ramses model was examined in an attempt to determine the composition and the types of technologies used in the model. This enabled the development of a model that uses similar technologies, but takes a fraction of the time to auto-code to reduce the turnaround time for experimentation. Lastly, the model was used to run a wide array of experiments and collect data to obtain knowledge about where to search for bottlenecks in the Ramses model. The resulting contributions of the overall effort consist of an experimental model for further investigation into the subject, as well as several automation tools to assist in analyzing the model, and a reference document offering insight to the auto-coding process, including documentation of the tools used in the model analysis, data illustrating some potential problem areas in the auto-coding process, and recommendations on areas or practices in the current Ramses model that should be further investigated. Several skills were required to be built up over the course of the internship project. First and foremost, my Simulink skills have improved drastically, as much of my experience had been modeling electronic circuits as opposed to software models. Furthermore, I am now comfortable working with the Simulink Auto-coder, a tool I had never used until this summer; this tool also tested my critical thinking and C++ knowledge as I had to interpret the C++ code it was generating and attempt to understand how the Simulink model affected the generated code. I had come into the internship with a solid understanding of Matlab code, but had done very little in using it to automate tasks, particularly Simulink tasks; along the same lines, I had rarely used shell script to automate and interface with programs, which I gained a fair amount of experience with this summer, including how to use regular expression. Lastly, soft-skills are an area everyone can continuously improve on; having never worked with NASA engineers, which to me seem to be a completely different breed than what I am used to (commercial electronic engineers), I learned to utilize the wealth of knowledge present at JSC. I wish I had come into the internship knowing exactly how helpful everyone in my branch would be, as I would have picked up on this sooner. I hope that having gained such a strong foundation in Simulink over this summer will open the opportunity to return to work on this project, or potentially other opportunities within the division. The idea of leaving a project I devoted ten weeks to is a hard one to cope with, so having the chance to pick up where I left off sounds appealing; alternatively, I am interested to see if there are any opening in the future that would allow me to work on a project that is more in-line with my research in estimation algorithms. Regardless, this summer has been a milestone in my professional career, and I hope this has started a long-term relationship between JSC and myself. I really enjoy the thought of building on my experience here over future summers while I work to complete my PhD at Missouri University of Science and Technology.

Gualdoni, Matthew J.↗
NASA NTRS2012

Small Fast Spectrum Reactor Designs Suitable for Direct Nuclear Thermal Propulsion

Advancement of U.S. scientific, security, and economic interests through a robust space exploration program requires high performance propulsion systems to support a variety of robotic and crewed missions beyond low Earth orbit. Past studies, in particular those in support of the Space Exploration Initiative (SEI), have shown nuclear thermal propulsion systems provide superior performance for high mass high propulsive delta-V missions. The recent NASA Design Reference Architecture (DRA) 5.0 Study re-examined mission, payload, and transportation system requirements for a human Mars landing mission in the post-2030 timeframe. Nuclear thermal propulsion was again identified as the preferred in-space transportation system. A common nuclear thermal propulsion stage with three 25,000-lbf thrust engines was used for all primary mission maneuvers. Moderately lower thrust engines may also have important roles. In particular, lower thrust engine designs demonstrating the critical technologies that are directly extensible to other thrust levels are attractive from a ground testing perspective. An extensive nuclear thermal rocket technology development effort was conducted from 1955-1973 under the Rover/NERVA Program. Both graphite and refractory metal alloy fuel types were pursued. Reactors and engines employing graphite based fuels were designed, built and ground tested. A number of fast spectrum reactor and engine designs employing refractory metal alloy fuel types were proposed and designed, but none were built. The Small Nuclear Rocket Engine (SNRE) was the last engine design studied by the Los Alamos National Laboratory during the program. At the time, this engine was a state-of-the-art graphite based fuel design incorporating lessons learned from the very successful technology development program. The SNRE was a nominal 16,000-lbf thrust engine originally intended for unmanned applications with relatively short engine operations and the engine and stage design were constrained to fit within the payload volume of the then planned space shuttle. The SNRE core design utilized hexagonal fuel elements and hexagonal structural support elements. The total number of elements can be varied to achieve engine designs of higher or lower thrust levels. Some variation in the ratio of fuel elements to structural elements is also possible. Options for SNRE-based engine designs in the 25,000-lbf thrust range were described in a recent (2010) Joint Propulsion Conference paper. The reported designs met or exceeded the performance characteristics baselined in the DRA 5.0 Study. Lower thrust SNRE-based designs were also described in a recent (2011) Joint Propulsion Conference paper. Recent activities have included parallel evaluation and design efforts on fast spectrum engines employing refractory metal alloy fuels. These efforts include evaluation of both heritage designs from the Argonne National Laboratory (ANL) and General Electric Company GE-710 Programs as well as more recent designs. Results are presented for a number of not-yet optimized fast spectrum engine options.

Schnitzler, Bruce G.↗