HPC container runtimes have minimal or no performance impact
Explore the source record for details and available documents.
SEARCH · Search NASA
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.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Finding where people live and the vulnerabilities of man-made facilities during natural disasters, is not only critical for rescue efforts, but also essential for damage assessment in the aftermath. Leveraging on the availability of high resolution satellite imagery, advances in machine learning and high performance computing hardware, it is now possible to generate geographical maps for man-made facilities at scale. Mapping from satellite imagery can be a daunting task due to the enormous amount of data to be processed over large areas. In this short paper we take advantage of annotated satellite imagery and automate the semantic labeling of mobile home parks using an efficient framework rooted in patch-based and pixel-level classification. This multilevel labeling effort is a precursor to our future goal for deploying very large scale deep convolutional neural networks toward both broad and finer characterization of man-made structures from one-meter resolution NAIP images.
In keeping with the trend of heterogeneity in high-performance computing, hardware manufacturers and vendors are developing new architectures and associated software stacks (e.g., libraries) to harness the best possible performance from commonly used kernels (e.g., linear algebra kernels). However, kernels tuned for one architecture are not portable to others. Moreover, the coexistence of different architectures in a single node makes orchestration difficult. To address these challenges, we introduce LaRIS, a portable framework for LAPACK functionalities. LaRIS ensures a separation between linear algebra algorithms and vendor-library kernels by using the IRIS run time and IRIS-BLAS library. Such abstraction at the algorithm level makes the implementation completely agnostic to the vendor library and architecture. LaRIS uses the IRIS run time to dynamically select the vendor-library kernel and suitable processor architecture at run time. Through LU factorization, we demonstrate that LaRIS can fully utilize different heterogeneous systems by launching and orchestrating different vendor-library kernels without any change in the source code.
This work considers the task of finding an AC-feasible operating point of a severely damaged transmission network while ensuring that a maximal amount of active power loads can be delivered. This AC Maximal Load Delivery (AC-MLD) task is a nonlinear, non-convex optimization problem, and is incredibly challenging to solve on large-scale transmission system datasets. This work demonstrates that convex relaxations of the AC-MLD problem provide a reliable and scalable method for finding high-quality bounds on the amount of active power that can be delivered in the AC-MLD problem. To demonstrate their effectiveness, the solution methods proposed in this work are rigorously evaluated on 1000 N – k scenarios on seven power networks ranging in size from 70 to 6000 buses. Furthermore, the most effective relaxation of the AC-MLD problem converges in less than 20 seconds on commodity computing hardware for all 7000 of the scenarios considered.
Deep learning (DL) models have received much attention lately for their ability to achieve expert-level performance on the accurate automated analysis of chest X-rays (CXRs). Recently available public CXR datasets include high resolution images, but state-of-the-art models are trained on reduced size images due to limitations on graphics processing unit memory and training time. As computing hardware continues to advance, it has become feasible to train deep convolutional neural networks on high-resolution images without sacrificing detail by downscaling. This study examines the effect of increased resolution on CXR classification performance. We used the publicly available MIMIC-CXR-JPG dataset, comprising 377,110 high resolution CXR images for this study. We applied image downscaling from native resolution to 2048 × 2048 pixels, 1024 × 1024 pixels, 512 × 512 pixels, and 256 × 256 pixels and then we used the DenseNet121 and EfficientNet-B4 DL models to evaluate clinical task performance using these four downscaled image resolutions. We find that while some clinical findings are more reliably labeled using high resolutions, many other findings are actually labeled better using downscaled inputs. We qualitatively verify that tasks requiring a large receptive field are better suited to downscaled low resolution input images, by inspecting effective receptive fields and class activation maps of trained models. Lastly, we show that stacking an ensemble across resolutions outperforms each individual learner at all input resolutions while providing interpretable scale weights, indicating that diverse information is extracted across resolutions.
Explore the source record for details and available documents.
Draco is an object-oriented component library geared towards numerically intensive, radiation (particle) transport applications built for parallel computing hardware. It consists of semi-independent packages and a robust build system. The packages in Draco provide a set of components that can be used by multiple clients to build transport codes. The build system can also be extracted for use in clients. Software includes smart pointers, Design-by-Contract assertions, unit test framework, wrapped MPI functions, a file parser, unstructured mesh data structures, a random number generator, root finders and an angular quadrature component.
This talk covers outlines the history of PaScalbb and then proceeds to outline testing that we need/want to do in order to design future HPC networks here at LANL.
As part of the NEAMS activity in DOE, a fast reactor simulation program was launched in April 2007, termed SHARP. The initial goal of SHARP was to develop a suite of modern simulation tools specifically for the analysis and design of sodium cooled fast reactors, and this goal has since been extended to consider all reactor types of interest. As part of that work, a high‐fidelity deterministic neutron transport code named UNIC was started, providing a common framework for fast reactor neutronics tools. Focusing on UNIC, spatial de‐homogenization increased the typical neutronics problem with a few hundred million degrees of freedom (DOF) into ones with hundreds of trillions of DOF. Such problems are beyond the capabilities of existing supercomputers. With the advent of future computer hardware, these high‐fidelity problems can be solved but only after a multiyear development program to adapt codes such as UNIC to that technology. UNIC can presently be used on smaller problems (single assembly, multiple assemblies) with today’s hardware, but it is also necessary to provide tools to solve the larger problems of interest, albeit without the aforementioned level of spatial de‐homogenization. To address the desire to demonstrate the coupling capability of NEAMS, part of the FY12 funding was focused on preparing the legacy codes, termed PROTEUS‐Fast, which are fully capable of solving full‐scale fast reactor problems. A bulk of the work this year was focused on preparing and distributing the PROTEUS‐Fast codes for use by all NEAMS users. The demonstration of a full coupling capability will require more follow on work, but significant progress was made this year preparing the interface coupling routines and preparing PROTEUS‐Fast for use. The primary targets of interest in PROTEUS‐Fast are DIF3D, DIF3D‐K (kinetics), and VARI3D (reactivity coefficients), with REBUS (fuel cycle) being an auxiliary component updated for convenience. All codes in PROTEUS‐Fast were verified and added to the BuildBot regression test. Unfortunately, all of the older codes suffered significant problems from the deprecated coding practices used within them and resulting conflicts with modern compilers. After some work, all of these issues were resolved and we feel confident in releasing them for use by others NEAMS members. With regard to specific upgrades, DIF3D‐VARIANT was upgraded with a few months of effort to allow the large scale problems of interest to be executed. After significant study, we choose not to modify VARI3D to include a transport option, but to develop a new code, PERSENT (partially funded by an external project) which uses the DIF3D‐VARIANT transport solution option. The outstanding features missing from MC 2 ‐3 were added, and MC 2 ‐3 has been released to RSICC. All of the remaining PROTEUS‐Fast codes are either finished with the copyright and licensing procedure or are actively going through it. While that process will delay the targeted August 2012 release to RSICC for a few more months, the codes are ready for immediate use and can be connected to the framework either by us or by others as the interface readers and documentation will be released with the PROTEUS‐Fast codes.
Abstract not provided.
Globally, floods are among the most frequent and costly natural disasters, both socially and economically. Many of these floods result from excess rainfall collecting in streams and rivers, then overtopping banks and flowing overland into urban environments, damaging critical infrastructure. In Fiscal Year 2012, the National Infrastructure Simulation and Analysis Center’s (NISAC) goal was to enhance its existing flood modeling and simulation capability to be able to estimate flooding from rainfall-runoff events. To accomplish this task, NISAC modified the existing flood modeling code-base using commodity high-performance computing hardware, making it more suitable for regional rainfall events. NISAC also implemented new modules within the existing flood code to account for hydrologic processes, such as spatially variable rainfall and infiltration. Analysts subjected the new hydrologic model to a verification and validation process, wherein analysts evaluated the model using a bench-scale experiment and watershed scale rainfall-runoff events. The bench-scale experiment indicated that the new hydrologic model is mass and momentum conserving and can adequately simulate runoff for depths of inches of water. The watershed scale simulations indicated that the hydrologic model is a good predictor of the discharge hydrograph when compared to observations if the input parameters, such as roughness and hydraulic conductivity, are adequately calibrated. The intended use of the hydrologic model is to estimate flood potential during extreme rainfall events. The accuracy of the predictions is largely dependent on the availability of accurate spatial data for soil and roughness parameters. In addition, each simulation should be accompanied with a calibration exercise, which requires historical rainfall and river discharge observations.
In this report, a project plan is developed, focused on industry applications, using Risk-Informed Safety Margin Characterization (RISMC) tools and methods applied to realistic, relevant, and current interest issues to the operating nuclear fleet. RISMC focuses on modernization of nuclear power safety analysis (tools, methods and data); implementing state-of-the-art modeling techniques (which include, for example, enabling incorporation of more detailed physics as they become available); taking advantage of modern computing hardware; and combining probabilistic and mechanistic analyses to enable a risk informed safety analysis process. The modernized tools will maintain the current high level of safety in our nuclear power plant fleet, while providing an improved understanding of safety margins and the critical parameters that affect them. Thus, the set of tools will provide information to inform decisions on plant modifications, refurbishments, and surveillance programs, while improving economics. This set of tools will also benefit the design of new reactors, enhancing safety per unit cost of a nuclear plant. The proposed plan will focus on application of the RISMC toolkit, in particular, solving realistic problems of important current issues to the nuclear industry, in collaboration with plant owners and operators to demonstrate the usefulness of these tools in decision making.