SEARCH · Engineering Papers
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.
Damselfly is a model-based parallel network simulator. It can simulate communication patterns of High Performance Computing applications on different network topologies. It outputs steady-state network traffic for a communication pattern, which can help in studying network congestion and its impact on performance.
Damselfly is a model-based parallel network simulator. It can simulate communication patterns of High Performance Computing applications on different network topologies. It outputs steady-state network traffic for a communication pattern, which can help in studying network congestion and its impact on performance.
This article considers the low-observability state estimation problem in power distribution networks and develops a decentralized state estimation algorithm leveraging the matrix completion methodology. Matrix completion has been shown to be an effective technique in state estimation that exploits the low dimensionality of the power system measurements to recover missing information. This technique can utilize an approximate (linear) load flow model, or it can be used with no physical models in a network where no information about the topology or line admittance is available. The direct application of matrix completion algorithms requires solving a semi-definite programming (SDP) problem, which becomes computationally challenging for large networks. We therefore develop a decentralized algorithm that capitalizes on the popular proximal alternating direction method of multipliers (proximal ADMM). The method allows us to distribute the computation among different areas of the network, leading to a scalable algorithm. By doing all computations at individual control areas and only communicating with neighboring areas, the algorithm eliminates the need for data to be sent to a central processing unit and thus increases efficiency and contributes to the goal of autonomous control of distribution networks. We illustrate the advantages of the proposed algorithm numerically using standard IEEE test cases.
The UC Davis team conducted architecture, algorithm and experimental studies for next generation ultra-high-bandwidth optical networks in support of extreme-scale science applications. In particular, over the course of this 3-year project, the UC Davis team achieved the following main accomplishments: Designed a network architecture for automated multi-domain service provisioning: Designed information exchange schemes between autonomous systems (domains) Designed application programming interfaces to orchestrate resource reservation in data centers, high performance computation facilities, and communication networks belonging to different autonomous systems Experimentally assessed the multi-domain network architecture in a distributed field trial set-up connecting premises in three continents Designed a broker-based orchestrated multi-domain optical networking framework supporting observe-analyze-act cycle based cognitive inter-domain service provisioning: Designed an inter-domain operation mechanism, e.g., topology abstraction, for the broker plane and domain managers to collaboratively accomplish inter-domain service provisioning while respecting the autonomy of each administrative domain Designed an alien wavelength monitoring scheme and an artificial neural network based optical signal-to-noise ratio estimator for quality-of-transmission aware inter-domain service provisioning Developed a deep neural network based multi-domain traffic predictor and a cognitive routing and spectrum assignment algorithm to realize knowledge-based autonomous traffic engineering in multi-domain networks Implemented an optical signal-to-noise ratio estimator and traffic predictor with Google’s Tensor flow and assessed their performance using experimental and synthetic data Demonstrated the benefits of the proposed cognitive schemes with experiments ona two-domain elastic optical network testbed and evaluations on a six-domain simulation framework Built a two-domain software-defined networking control plane testbed using the ONOS platform and demonstrated the workflow of broker and domain manager cooperation, including hierarchical performance monitoring, topology abstraction and cognitive resource allocation. Designed self-learning cognitive service provisioning schemes using advanced machine learning technologies: Designed a self-taught anomaly detection framework for optical networks, including an unsupervised data clustering module for pattern recognition, a supervised data regression and classification module for online anomaly detection, and the system operation principle Experimentally validated the effectiveness of the self-taught anomaly detection framework with various failure/anomaly scenarios. Designed a self-learning autonomic routing, modulation and spectrum assignment (RMSA) provisioning framework for elastic optical networks with deep reinforcement learning. Designed effective modeling and training mechanisms for the self-learning RMSA framework Conducted extensive numerical simulations with various traffic patterns and topologies to verify the advantage of the self-learning RMSA framework.
Methods, devices, and systems for GPU cache injection. A GPU compute node includes a network interface controller (NIC) which includes NIC receiver circuitry which can receive data for processing on the GPU, NIC transmitter circuitry which can send the data to a main memory of the GPU compute node and which can send coherence information to a coherence directory of the GPU compute node based on the data. The GPU compute node also includes a GPU which includes GPU receiver circuitry which can receive the coherence information; GPU processing circuitry which can determine, based on the coherence information, whether the data satisfies a heuristic; and GPU loading circuitry which can load the data into a cache of the GPU from the main memory if on the data satisfies the heuristic.
Here, FPGA-based heterogeneous computing platform, due to its extreme logic reconfigurability, emerges to be a strong contender as computing fabric in modern AI. As a result, various FPGA-based accelerators for deep CNN—the key driver of modern AI—have been proposed due to their advantages of high performance, reconfigurability, and fast development round, etc. In general, the consensus among researchers is that, although FPGA-based accelerator can achieve much higher energy efficiency, its raw computing performance lags behind when compared with GPUs with similar logic density. In this paper, we develop an alternative methodology to efficiently implement CNNs with FPGAs that outperform GPUs in terms of both power consumption and performance. Our key idea is to design a scalable hardware architecture and circuit design for large-scale CNNs that leverages a stochastic-based computing principle. Specifically, there are three major performance advantages. First, all key components of our deep learning CNN are designed and implemented to compute stochastically, thus achieving excellent computing performance and energy efficiency. Second, because our proposed CNN architecture enables a stream-mode computing, all of its stages can process even the partial results from preceding stages, therefore not incurring unnecessary latency due to data dependency. Finally, our FPGA-based deep CNN also provides a superior hardware scalability when compared with conventional FPGA implementations by reducing the bandwidth requirement between layers. The results show that our proposed CNN architecture significantly outperforms all previous FPGA-based deep CNN implementation approaches. It achieves 1.58x more GOPS, 6.42x more GOPS/Slice, and 10.92x more GOPS/W when compared with state-of-the-art CNN architecture. The top-5 accuracy of stochastic VGG-16 CNN is 86.77 percent with 18.91 fps frame rate.
The focus of this milestone was to implement and test new algorithms and optimizations for improved GPU performance by molecular dynamics (MD) and related particle codes. The work was done in the CoPA Cabana library, the CabanaMD mini-app which uses Cabana to implement and benchmark prototypical MD models, and the LAMMPS MD code which is used by the ECP EXAALT project as well as for materials modeling generally. Efforts included (a) implementation of new or optimized algorithms for fundamental MD kernels like neighbor-list building and FFTs which are also widely used in other particle methods, (b) performance testing of data structure layouts for both simple pairwise and computationally intense neural network potentials, (c) optimizations aimed at improving GPU performance for models with small atom counts, and (d) optimizations specific to the SNAP interatomic potential due to its importance for the EXAALT project.
Abstract The sequence‐specific hybridization of DNA facilitates its use as a building block for designer nanoscale structures and reaction networks that perform computations. However, the strong binding energy of Watson–Crick base pairing that underlies this specificity also causes the DNA dehybridization rate to depend sensitively on sequence length and temperature. This strong dependency imposes stringent constraints on the design of multi‐step DNA reactions. Here we show how an ATP‐dependent helicase, Rep‐X, can drive specific dehybridization reactions at rates independent of sequence length, removing the constraints of equilibrium on DNA hybridization and dehybridization. To illustrate how this new capacity can speed up designed DNA reaction networks, we show that Rep‐X extends the range of conditions where the primer exchange reaction, which catalytically adds a domain provided by a hairpin template to a DNA substrate, proceeds rapidly.
Abstract The sequence‐specific hybridization of DNA facilitates its use as a building block for designer nanoscale structures and reaction networks that perform computations. However, the strong binding energy of Watson–Crick base pairing that underlies this specificity also causes the DNA dehybridization rate to depend sensitively on sequence length and temperature. This strong dependency imposes stringent constraints on the design of multi‐step DNA reactions. Here we show how an ATP‐dependent helicase, Rep‐X, can drive specific dehybridization reactions at rates independent of sequence length, removing the constraints of equilibrium on DNA hybridization and dehybridization. To illustrate how this new capacity can speed up designed DNA reaction networks, we show that Rep‐X extends the range of conditions where the primer exchange reaction, which catalytically adds a domain provided by a hairpin template to a DNA substrate, proceeds rapidly.
Truck platooning enables a group of trucks to move close together, which helps reduce truck fuel use and increase effective road capacity. In this paper, a system-level equilibrium model is developed to characterize spontaneous truck platooning with coexistence of non-platooning vehicles in a network, by explicitly accounting for the interlocking relationship among platoon formation time, truck fuel saving, and increase in effective road capacity. To equilibrate the relationships, an algorithm is proposed which involves a diagonalization approach and a bush based algorithm to solve decomposed subproblems. The condition of proportionality is imposed to obtain unique traffic flows for each class of vehicles on road links. In addition, a spatially constrained multivariate clustering technique is employed to construct origin/destination zones that are smaller than the coarse Freight Analysis Framework (FAF) zones, while maintaining reasonable computational burden for network traffic assignment. Model implementation in the U.S. shows that platooning could lead to 7.9% fuel saving among platoonable trucks in 2025 and a comparable increase in effective capacity of platoonable road links, which would account for 60% of rural interstate roads. The fuel saving and road capacity improvement translate into an annual cost reduction of $\$$868 million for the U.S. intercity trucking sector and reduced road infrastructure investment needs worth $\$$4.8 billion. Extensive sensitivity analysis further reveals that fuel saving of platoonable trucks increases with platoon size but decreases with inter-truck distance in a platoon. Fuel saving potential suggests that priority should be given to rural rather than urban roads in deploying platooning technologies. As we expected, greater market penetration of platooning technologies means higher fuel saving and greater increase in effective road capacity.
Jobs on most high-performance computing (HPC) systems share the network with other concurrently executing jobs. Network sharing leads to contention that can severely degrade performance. Here we investigate the use of Quality of Service (QoS) mechanisms to reduce the negative impacts of network contention. QoS allows users to manage resource sharing between network flows and to provide bandwidth guarantees to specific flows. Our results show that careful use of QoS reduces the impact of network contention for specific jobs, resulting in up to a 40% performance improvement. In some cases, it completely eliminates the impact of contention. It achieves these improvements with limited negative impact to other jobs; any job that experiences performance loss typically degrades less than 5%, and often much less. Our approach can help ensure that HPC machines maintain high levels of throughput as per-node compute power continues to increase faster than network bandwidth.
The Assistant Secretary of the Army (Installations, Energy and Environment) tasked the Pacific Northwest National Laboratory to examine plug loads in typical Army buildings. Plug loads (also known as miscellaneous electric loads (MELs)) represent the electricity used by appliances and devices that are plugged in or hardwired and serve functions outside of a building’s core end uses. Common plug loads include computers, printers, copiers, networking devices, refrigerators, and vending machines. They also include personal electronic devices such as televisions, smart phones, tablets, and gaming systems. Examples of hardwired MELs include elevators, air compressors, and fire and security systems. The findings from this study confirm that significant energy is consumed within Army buildings by plug load devices and hardwired MEL equipment. A number of opportunities are identified for reducing unnecessary energy use that could save the Army over $5 million per year when broadly applied. Army regulations clearly spell out expectations for the purchase and operation of information technology equipment (computers, laptops, monitors, printers, and multi-function devices). However, the policies regarding the shutdown or activation of sleep and other lower power modes after 30 minutes of inactivity (15 minutes for monitors) do not appear to be consistently followed. There are many effective approaches and pathways for impacting change as it relates to improving awareness, implementing measures, and adjusting behaviors to identify and reduce plug load energy use. The Army should prioritize and consider deploying all of these to better understand and manage plug load equipment to save energy and enhance resilience across their facilities. Engaging the building occupants who use these devices daily via outreach and education should be a strong component of the strategy. The focus should be on reducing waste without sacrificing productivity or the benefits that many of these devices provide. Continued evaluation of plug loads beyond that performed here is important to gather lessons from additional building and equipment types, and to stay aware of evolving device technology and management options. This will highlight additional needs for policies, best practices, control technologies, and education of personnel to achieve real reductions in energy waste from plug load equipment. It is recommended that this study may serve as the foundation for a broader and sustained focus on plug loads and MELs, towards simultaneously enhancing the productivity, readiness, and resilience of the Army while reducing energy use and demand, and freeing up resources to better support the mission.
The complexity of scientific pursuits is increasing rapidly with aspects that require dynamic integration of experiment, observation, theory, modeling, simulation, visualization, machine learning (ML), artificial intelligence (AI), and analysis. Research projects across the Department of Energy (DOE) are increasingly data and compute intensive. Innovative research teams are accelerating the pace of discovery by using high-performance computational and data tools in their research workflows and leveraging multiple research infrastructures. Additionally, several recent high-level U.S. government reports underscore the necessity of a new advanced computing ecosystem for international competitiveness and national security. International competitors are moving forward with major research infrastructure integration efforts that seek to capture a competitive advantage in the global innovation race. Owing to its unparalleled constellation of world-class experimental and observational facilities and high-performance and extreme-scale computational, data, and networking infrastructure, DOE is positioned to be a global leader in this new era of integrated science. However, this new integration paradigm will demand continuing evolution to ensure the U.S. remains a global leader in research and innovation. The DOE Office of Science (SC) has seized on the strategic importance of integration and has adopted a vision for Integrated Research Infrastructure (IRI): To empower researchers to meld DOE’s world-class research tools, infrastructure, and user facilities seamlessly and securely in novel ways to radically accelerate discovery and innovation. To respond to the evolving computational requirements of research and the competitive international innovation landscape, experimental facilities could be connected with high performance computing resources for near real-time analysis, and resources should be provided for merging enormous and diverse data for AI/ML techniques and analysis.
Knowledge distillation is a form of model compression that allows artificial neural networks of different sizes to learn from one another. Its main application is the compactification of large deep neural networks to free up computational resources, in particular on edge devices. In this article, we consider proton-proton collisions at the High-Luminosity Large Hadron Collider (HL-LHC) and demonstrate a successful knowledge transfer from an event-level graph neural network (GNN) to a particle-level small deep neural network (DNN). Our algorithm, DistillNet, is a DNN that is trained to learn about the provenance of particles, as provided by the soft labels that are the GNN outputs, to predict whether or not a particle originates from the primary interaction vertex. The results indicate that for this problem, which is one of the main challenges at the HL-LHC, there is minimal loss during the transfer of knowledge to the small student network, while improving significantly the computational resource needs compared to the teacher. This is demonstrated for the distilled student network on a CPU, as well as for a quantized and pruned student network deployed on a field programmable gate array. Our study proves that knowledge transfer between networks of different complexity can be used for fast artificial intelligence (AI) in high-energy physics that improves the expressiveness of observables over non-AI-based reconstruction algorithms. Such an approach can become essential at the HL-LHC experiments, e.g. to comply with the resource budget of their trigger stages.
This final report is presented by Langston University (LU) for the project entitled "Langston University High Energy Physics" (LUHEP) under the direction of principal investigator (PI) and project director Professor Joel Snow. The project encompassed high energy physics research performed at hadron colliders. The PI is a collaborator on the DZero experiment at Fermi National Accelerator Laboratory in Batavia, IL, USA and the ATLAS experiment at CERN in Geneva, Switzerland and was during the entire project period from April 1, 1999 until May 14, 2012. Both experiments seek to understand the fundamental constituents of the physical universe and the forces that govern their interactions. In 1999 as member of the Online Systems group for Run 2 the PI developed a cross-platform Python-based, Graphical User Interface (GUI) application for monitoring and control of EPICS based devices for control room use. This served as a model for other developers to enhance and build on for further monitoring and control tasks written in Python. Subsequently the PI created and developed a cross-platform C++ GUI utilizing a networked client-server paradigm and based on ROOT, the object oriented analysis framework from CERN. The GUI served as a user interface to the Examine tasks running in the D\O\ control room which monitored the status and integrity of data taking for Run 2. The PI developed the histogram server/control interface to the GUI client for the EXAMINE processes. The histogram server was built from the ROOT framework and was integrated into the D\O\ framework used for online monitoring programs and offline analysis. The PI developed the first implementation of displaying histograms dynamically generated by ROOT in a Web Browser. The PI's work resulted in several talks and papers at international conferences and workshops. The PI established computing software infrastructure at LU and U. Oklahoma (OU) to do analysis of DZero production data and produce simulation data for the experiment. Eventually this included the FNAL SAM data grid system, the SAMGrid (SG) infrastructure, and the Open Science Grid software stacks for computing and storage elements. At the end of 2003 Snow took on the role of global Monte Carlo production coordinator for the DØ experiment. A role which continues til this day. In January of 2004 Snow started working with the SAMGrid development team to help debug, deploy, and integrate SAMGrid with DØ Monte Carlo production. Snow installed and configured SG execution and client sites at LUHEP and OUHEP, and a SG scheduler site at LUHEP. The PI developed a python based GUI (DAJ) that acts as a front end for job submission to SAMGrid. The GUI interfaces to the DZero Mone Carlo (MC) request system that uses SAM to manage MC requests by the physics analysis groups. DAJ significantly simplified SG job submission and was deployed in DZero in an effort to increase the user base of SG. The following year was the advent of SAMGrid job submission to the Open Science Grid (OSG) and LHC Computing Grid (LCG) through a forwarding mechanism. The PI oversaw the integration of these grids into the existing production infrastructure. The PI developed an automatic MC (Automc) request processing system capable of operating without user intervention (other than getting grid credentials), and able to submit to any number of sites on various grids. The system manages production at all but 2 sites. The system was deployed at Fermilab and remains operating there today. The PI's work in distributed computing resulted in several talks at international conferences. UTA, OU, and LU were chosen as the collaborating institutions that form the Southwest Tier 2 Center (SWT2) for ATLAS. During the project period the PI contributed to the online and offline software infrastructure through his work with the Run 2 online group, and played a major role in Monte Carlo production for DZero. During the part of the project period in which the PI served as MC production coordinator MC production increased very significantly. In the first year of the PI's tenure as production coordinator production was 159M events and 6.7~TB of data. During the last year of the project period production was 2,342~M events and 262~TB of data. That is a factor of 15 increase in events and 39 in data volume. The increase occurred with improvements in computer hardware and networks, through the use of grid technology on diverse resources, and through increased automation and efficiency of the production process. LU HEP developed and deployed the automatic MC request processing system in use at FNAL. The complementary strategies of automation and grid production served DZero well. Fermilab has recognized LU HEP's contribution to DZero by allowing the PI to devote full time to research activities by appointing him a guest scientist for the last six years of the project period.
Inspired by the connectivity mechanisms in the brain, neuromorphic computing architectures model Spiking Neural Networks (SNNs) in silicon. As such, neuromorphic architectures are designed and developed with the goal of having small, low power chips that can perform control and machine learning tasks. However, the power consumption of the developed hardware can greatly depend on the size of the network that is being evaluated on the chip. Furthermore, the accuracy of a trained SNN that is evaluated on chip can change due to voltage and current variations in the hardware that perturb the learned weights of the network. While efforts are made on the hardware side to minimize those perturbations, a software based strategy to make the deployed networks more resilient can help further alleviate that issue. In this work, we study Spiking Neural Networks in two neuromorphic architecture implementations with the goal of decreasing their size, while at the same time increasing their resiliency to hardware faults. We leverage an evolutionary algorithm to train the SNNs and propose a multiobjective fitness function to optimize the size and resiliency of the SNN. We demonstrate that this strategy leads to well-performing, small-sized networks that are more resilient to hardware faults.
Power and energy concerns are motivating chip manufacturers to consider future hybrid-core processor designs that may combine a small number of traditional cores optimized for single-thread performance with a large number of simpler cores optimized for throughput performance. This trend is likely to impact the way in which compute resources for network protocol processing functions are allocated and managed. In particular, the performance of MPI match processing is critical to achieving high message throughput. In this paper, we analyze the ability of simple and more complex cores to perform MPI matching operations for various scenarios in order to gain insight into how MPI implementations for future hybrid-core processors should be designed.