Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Benchmark testing”

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 811 records · Page 45
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

HipBone: A performance-portable graphics processing unit-accelerated C++ version of the NekBone benchmark

We present hipBone, an open-source performance-portable proxy application for the Nek5000 (and NekRS) computational fluid dynamics applications. HipBone is a fully GPU-accelerated C++ implementation of the original NekBone CPU proxy application with several novel algorithmic and implementation improvements which optimize its performance on modern fine-grain parallel GPU accelerators. Our optimizations include a conversion to store the degrees of freedom of the problem in assembled form in order to reduce the amount of data moved during the main iteration and a portable implementation of the main Poisson operator kernel. We demonstrate near-roofline performance of the operator kernel on three different modern GPU accelerators from two different vendors. We present a novel algorithm for splitting the application of the Poisson operator on GPUs which aggressively hides MPI communication required for both halo exchange and assembly. Our implementation of nearest-neighbor MPI communication then leverages several different routing algorithms and GPU-Direct RDMA capabilities, when available, which improves scalability of the benchmark. We demonstrate the performance of hipBone on three different clusters housed at Oak Ridge National Laboratory, namely, the Summit supercomputer and the Frontier early-access clusters, Spock and Crusher. Our tests demonstrate both portability across different clusters and very good scaling efficiency, especially on large problems.

Computer Science↗
DOE OSTI2026

Robotic automation of maintenance work in nuclear power plants a cross-sector survey and roadmap

Nuclear power plants face increasing cost pressures, workforce constraints (aging workforce and skilled labor shortages), and safety requirements that are accelerating interest in robotic systems for inspection and maintenance. We conducted semi-structured interviews with personnel from seven U.S. nuclear utilities and compared deployment models, operational use cases, and integration practices with those reported by participants in the oil, gas, and petrochemical sector. In nuclear plants, robotic use remains concentrated in inspection—particularly indoor unmanned aerial vehicles and submersible remotely operated vehicles—with limited application to physical maintenance tasks. Reported near-term value includes reduced radiological and industrial risk, reduced outage labor, and improved data for planning and condition assessment. Key barriers include integration and data-interoperability constraints, operator qualification requirements, cybersecurity review burden, and difficulty demonstrating reliability in plant-representative environments. Cross-sector benchmarking highlights organizational and deployment practices that may help nuclear plants scale from pilots to routine use. We propose a deployment-oriented roadmap emphasizing modular payload strategies, representative qualification pathways and testing environments, and improved data governance to support safe and economically justified expansion of robotics in operating nuclear power plants.

11 - NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2026

MIP dQ/dx Calibration in DUNE ND-LAr Prototypes with Pixelated Charge Readout

The Deep Underground Neutrino Experiment (DUNE) will be a next-generation long baseline neutrino oscillation experiment that will employ LArTPC technology in a near detector placed at Fermilab and a far detector at the Sanford Underground Research Facility, at a baseline of 1300 km. The DUNE Liquid Argon Near Detector (ND-LAr) design takes into account the high neutrino intensity expected from the beam at the Long-Baseline Neutrino Facility (LBNF): 35 modules, each containing two optically separated time projection chambers, are instrumented with a pixel-based, true 3D charge readout alongside scintillation light traps to disentangle the O(100) interactions expected per 10us beam spill. A robust prototyping program supports ND-LAr’s design: the 2x2 Demonstrator consists of four scaled-down ND-LAr modules exposed to the NuMI beam at Fermilab, and the Full Scale Demonstrator (FSD) is a single ND-LAr module tested with cosmic rays at the University of Bern. We present here an analysis of minimum ionizing particle (MIP) tracks selected from 2x2 beam data and FSD cosmic ray data used to benchmark the pixel-based charge readout simulation and calibration in both detectors, with the ultimate goal of validating the design of DUNE ND-LAr as well as informing future calibration methods. This poster will showcase the dependence of the charge response on track inclination as well as a per-pixel dQ/dX extraction.

Mandujano, Roberto [UC, Irvine]↗
DOE OSTI2026

A Centralized AI Lakehouse Framework for Brain Tumor MRI Classification and Segmentation, University KPI Forecasting, and Water Potability Prediction

In many university and healthcare projects, models are built for very different data types such as tables, institutional time series, and medical images, but they are deployed as separate applications. In this work, that separation made testing and maintenance difficult because each module had its own pipeline and runtime requirements. This paper presents an integrated AI lakehouse-style implementation that runs three model pipelines inside one containerized backend. For medical imaging, we used MRI datasets from IEEE DataPort: a four-class classification set with 7012 images (5708 train/1304 test) and a segmentation set with 3063 image–mask pairs. The classification model (ResNet50 transfer learning) is evaluated using a proper train–validation–test protocol across multiple splits (80/10/10, 70/10/20, 60/10/30, and 10/30/60), achieving a test accuracy of 99.00% under the standard 80/10/10 split. Additionally, a patient-level evaluation is conducted using an external glioma dataset to provide a more realistic assessment without data leakage. The segmentation model (DeepLabV3-ResNet50) achieved 83.09% validation mIoU and 88.79% Dice score. For university KPI forecasting, we used annual IPEDS and NSF HERD data from 2010 to 2023 for three universities (BSU, EOU, and UAB). To examine the effect of preprocessing on forecasting performance, two case studies are conducted. In the first case, linear interpolation is applied to generate semester-level data. In the second case, the original annual data is used directly without interpolation. Random Forest regression and ARIMA models are evaluated using MAE, RMSE, MAPE, and R 2 . The results showed that interpolation improved apparent forecasting performance due to smoothing, while evaluation on the original annual data provided a more realistic assessment of model behavior. To further validate the framework on a larger dataset, an additional case study is conducted using a student dropout dataset. For water potability, we trained and compared multiple tabular classifiers on a large dataset (1,048,575 samples). A Random Forest model (100 trees, max depth 10) achieved 85.86% test accuracy and high recall for unsafe samples (0.8447). All modules are served via FastAPI and deployed together using Docker, with workflow automation routing requests to the correct endpoint. System-level benchmarking indicates that the backend maintains stable throughput and latency under concurrent requests.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Development of an Improved RELAP5-3D Model for the High Temperature Test Facility

High-temperature gas-cooled reactors (HTGRs) are rapidly approaching deployment. Confidence in transient analysis of these systems for design, optimization, and licensing calculations requires modeling and simulation tools that have been validated against data relevant to HTGR conditions. The High Temperature Test Facility (HTTF) is an integral effects thermal hydraulics test facility for prismatic HTGRs. In spring and summer of 2019, HTTF was used for a series of experiments that now serve as the basis for the OECD/NEA Thermal Hydraulic Code Validation Benchmark for High Temperature Gas-Cooled Reactors using HTTF Data (HTGR T/H Benchmark). This benchmark contains problems for systems code, computational fluid dynamics (CFD), and coupled systems code/CFD modeling representing lower plenum mixing and both the depressurized and pressurized conduction cooldown (DCC and PCC respectively) transients. Benchmark problems include exercises for code-to-code and code-to-data comparisons as well as an exercise for error scaling between HTTF and the Modular High Temperature Gas-Cooled Reactor, which serves as the basis for the HTTF design. Previous analysis as part of the HTGR T/H benchmark used a RELAP5-3D model developed at Idaho National Laboratory (INL) and demonstrated an ability to reproduce trends in the measured data but difficulties reproducing experimental values within their uncertainty. These difficulties were largely attributed to assumptions made during the development of the initial RELAP5-3D model, which predated the HTTF experiments. A significant cause of difficulty reproducing the measured temperatures may be the radial nodalization of the previous RELAP5-3D model. The new model provides a finer nodalization to assess the impact of radial nodalization and allows for asymmetric heating within the core, which was a feature of multiple HTTF experiments. In this paper, we present the new RELAP5-3D model of HTTF. In addition to describing the new model, this paper compares the new and old models and provides results for a full-power steady state, a DCC, and a PCC in HTTF. These analyses are based on the code-to-code comparison exercises for the DCC and PCC problems of the HTGR T/H benchmark. We present the results of these exercises from the new model and compare them to the results of the old model.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2025

Validation of Numerical Tools for Calculating Reactivity Feedback in Sodium Fast Reactors Using SEFOR Experimental Data

The Southwest Experimental Fast Oxide Reactor (SEFOR) was an experimental sodium-cooled fast breeder reactor operated from 1969 to 1972 with experiments designed to measure Doppler reactivity feedback in a wide temperature range from around 350 °F to temperatures approaching the melting point of mixed oxide fuel of around 5000 °F, providing valuable data for code validations. Co-supported by the Department of Energy (DOE) Fast Reactor Program (FRP) and the DOE Nuclear Energy Advanced Modeling and Simulation (NEAMS) program, the SEFOR benchmark project focused on using the experimental data to validate numerical tools that are used in industry and academia to design and license sodium-cooled fast reactors (SFRs). By the end of FY-25, substantial progress was achieved in the SEFOR benchmark study. A variety of numerical tools commonly used for modeling SFRs were applied to develop models for SEFOR core configurations I-D, I-E, I-I, and I-J. These included Monte Carlo codes such as MCNP, Serpent, and Shift; deterministic codes such as the legacy Argonne Reactor Computation (ARC) suite and the high-fidelity NEAMS code Griffin; and the system analysis code SAS4A/SASSYS-1 (SAS). Using these models, both SEFOR zero-power experiments and power-ascending tests were successfully simulated. Comparisons were performed against experimental measurements of core criticalities, reflector worth, kinetics parameters (Λ/βeff), isothermal reactivity feedback (from 350 °F to 760 °F at zero power), and power-ascending reactivity feedback (as power increased from 0.4 MW to 17 MW). In general, these comparisons demonstrated very good agreement between numerical results and experimental data. In Fiscal Year 26 (FY-26), the SEFOR benchmark project will continue to address the modeling issues identified in FY-25. Effort will focus on the simulation of reactivity insertion transients in SEFOR core II using the ARC/SAS model. Future work will also focus on incorporating BISON into the SEFOR core modeling process to enable the first Multiphysics simulations of the isothermal tests based on the MOOSE framework.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗
DOE OSTI2025

Phenomenology of electroweak portal dark showers: high energy direct probes and low energy complementarity

We investigate the phenomenology of a dark QCD sector interacting with the Standard Model (SM) via the electroweak (EW) portals. The portal interactions allow SM bosons, such as Z and h, or additional bosons that mix with them, to decay into dark quarks, producing dark showers. The light dark mesons are expected to be long-lived particles (LLPs), as their decays back to the SM states through the EW-portal interactions typically have macroscopic decay lengths. We focus on dark shower events initiated by various bosons at the Large Hadron Collider (LHC). The most prominent signal is the displaced decay of GeV-scale dark pions as LLPs. Current limits on dark shower signals at LHC detectors are recast from public data to provide simplified limits insensitive to UV physics details. Future limits in the high-luminosity phase and proposed auxiliary detectors are also projected. Additionally, we study the flavor-changing neutral current (FCNC) B decays into dark pions, obtaining both current and projected constraints at the LHC and other facilities. These constraints can be combined for specific models, which are illustrated in two EW-portal benchmarks: one with the heavy doublet fermion mediation and another with the Z′ mediator including a mass mixing. The collider reach shows significant potential to probe the parameter space unconstrained by EW precision tests, highlighting the necessity of dedicated LLP search strategies and facilities.

Dark Matter at Colliders↗
DOE OSTI2020

Build Orientation Dependent Microstructure in Polymer Laser Sintering: Relationship to Part Performance and Evolution with Aging

Laser sintering (LS) is widely used to produce functional polymeric parts; however, the resulting parts are often limited by their porous structure, and performance of the part may be strongly anisotropic. Relating the structural features of parts to build process conditions or powder feedstock has been explored previously. In contrast, little is known regarding how the unique internal structures of LS parts evolve and relate to their performance later in their life cycle, for example, after use in a real-world operating environment. In this study, a tightly controlled LS build process and measurement campaign by standard X-ray computed tomography (XCT), supported by high-resolution synchrotron XCT, was used to benchmark the internal microstructure (e.g. porosity) as a function of four different build orientations in printed polyamide-12 (PA-12) parts. The initial performance of the parts after printing was characterized by tensile testing and dynamic mechanical analysis (DMA). Arrays of PA-12 parts in different build orientations were exposed to steam (under pressure) and air across multiple temperature and time points for the purposes of accelerated aging through oxidative and hydrolytic breakdown. Exposing parts to steam dramatically altered the internal microstructure and functional properties such as the glass transition temperature, tensile properties, and damping behavior. Notably, post-aging XCT revealed large microstructural changes after only 40 h of steam treatment relative to the as-printed specimens. Furthermore, pore reorganization and changes in crystallinity occurred regardless of whether the steam temperature was aggressive enough to induce a significant loss in mechanical properties. The build orientation dependence on the aging rate was minimal with the relative trend in performance persisting across most metrics among the different orientations even after aging.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Code Verification and Solution Verification framework in pin-resolved neutron transport code MPACT

Program verification in scientific computing encompasses the application of formal and mathematical techniques to a scientific computing code for its credibility, accuracy, and validity. Code Verification identifies bugs and performance issues in the software development stage. Solution Verification assesses the applicability of the code and the accuracy of the solution to problems of interest. Both activities utilize application cases and quantify the error against prescribed acceptance criteria. However, simply executing more application cases does not guarantee stronger or more comprehensive credibility. Here, we establish a verification framework that involves Code Verification and Solution Verification, both of which work together such that the overarching goal of “converge to the correct answer for the intended application” can be reasonably inferred. The application of such a verification framework is demonstrated using the pin-resolved neutron transport code MPACT, where standard unit tests and regression tests are covered, and where the Method of Exact Solutions and the Method of Manufactured Solutions are successfully used. Additionally, the applicability of Method of Manufactured Solutions is extended to the OECD/NEA C5G7 benchmark problems of practical material and geometric configurations. Solution Verification activities are demonstrated on a practical hierarchy of application models of increasing complexity ranging from 2D pin cell problems to 3D assembly problems. The convergence behavior and rate of convergence with respect to each individual variable are studied and provided. This framework can be adapted broadly to other fields involving scientific computing codes.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

High-order finite element method for atomic structure calculations

We introduce featom, an open source code that implements a high-order finite element solver for the radial Schrödinger, Dirac, and Kohn-Sham equations. The formulation accommodates various mesh types, such as uniform or exponential, and the convergence can be systematically controlled by increasing the number and/or polynomial order of the finite element basis functions. The Dirac equation is solved using a squared Hamiltonian approach to eliminate spurious states. Here, to address the slow convergence of the $κ=±1$ states due to divergent derivatives at the origin, we incorporate known asymptotic forms into the solutions. We achieve a high level of accuracy (10 -8 Hartree) for total energies and eigenvalues of heavy atoms such as uranium in both Schrödinger and Dirac Kohn-Sham solutions. We provide detailed convergence studies and computational parameters required to attain commonly required accuracies. Finally, we compare our results with known analytic results as well as the results of other methods. In particular, we calculate benchmark results for atomic numbers (Z) from 1 to 92, verifying current benchmarks. We demonstrate significant speedup compared to the state-of-the-art shooting solver dftatom. An efficient, modular Fortran 2008 implementation, is provided under an open source, permissive license, including examples and tests, wherein particular emphasis is placed on the independence (no global variables), reusability, and generality of the individual routines.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2026

ENDF/B-VIII.1: Updated Nuclear Reaction Data Library for Science and Applications

The ENDF/B-VIII.1 library is the newest recommended evaluated nuclear data file by the Cross Section Evaluation Working Group (CSEWG) for use in nuclear science and technology applications, and incorporates advances made in the six years since the release of ENDF/B-VIII.0. Among key advances made are that the 239 Pu file was reevaluated by a joint international effort and that updated 16,18 O, 19 F, 28–30 Si, 50–54 Cr, 55 Mn, 54,56,57 Fe, 63,65 Cu, 139 La, 233,235,238 U, and 240,241 Pu neutron nuclear data from the IAEA coordinated INDEN collaboration were adopted. Over 60 neutron dosimetry cross sections were adopted from the IAEA's IRDFF-II library. In addition, the new library includes significant changes for 3 He, 6 Li, 9 Be, 51 V, 88 Sr, 103 Rh, 140,142 Ce, Dy, 181 Ta, Pt, 206–208 Pb, and 234,236 U neutron data, and new nuclear data for the photonuclear, charged-particle and atomic sublibraries. Numerous thermal neutron scattering kernels were reevaluated or provided for the very first time. On the covariance side, work was undertaken to introduce better uncertainty quantification standards and testing for nuclear data covariances. The significant effort to reevaluate important nuclides has reduced bias in the simulations of many integral experiments with particular progress noted for fluorine, copper, and stainless steel containing benchmarks. Data issues hindered the successful deployment of the previous ENDF/B-VIII.0 for commercial nuclear power applications in high burnup situations. These issues were addressed by improving the 238 U and 239,240,241 Pu evaluated data in the resonance region. The new library performance as a function of burnup is similar to the reference ENDF/B-VII.1 library.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2024

Top-down mass spectrometry of native proteoforms and their complexes: a community study

The combination of native electrospray ionization with top-down fragmentation in mass spectrometry (MS) allows simultaneous determination of the stoichiometry of noncovalent complexes and identification of their component proteoforms and cofactors. Although this approach is powerful, both native MS and top-down MS are not yet well standardized, and only a limited number of laboratories regularly carry out this type of research. Further, to address this challenge, the Consortium for Top-Down Proteomics initiated a study to develop and test protocols for native MS combined with top-down fragmentation of proteins and protein complexes across 11 instruments in nine laboratories. Here we report the summary of the outcomes to provide robust benchmarks and a valuable entry point for the scientific community.

47 OTHER INSTRUMENTATION↗
DOE OSTI2021

A modified many-body dissipative particle dynamics model for mesoscopic fluid simulation: methodology, calibration, and application for hydrocarbon and water

The many-body dissipative particle dynamics (mDPD) is a prominent mesoscopic multiphase model for fluid transport in mesoconfinement. However, it has been a long-standing challenge for mDPD (and other multiphase-enabled DPD models) to accurately predict real-fluid static and dynamic properties simultaneously. We have developed a modified mDPD model to overcome the issue and a rigorous calibration approach that uses reference data, including experimental and/or molecular dynamics (MD) simulations to parameterise the modified mDPD for real fluids. We choose heptane as a representative example of hydrocarbon in source rocks to demonstrate the model's capability to accurately predict the equation of state (EOS), free surface tension, diffusivity, and viscosity. Our timing test shows that the modified mDPD is 400–500 times faster than its MD counterpart for simulating bulk heptane in equivalent volumes. To further demonstrate the robustness of the model, we revisited the benchmark problem of mesoscopic modelling of water, in which all the previous works on DPD reported only a limited portion of the water properties. Here, we show that the modified mDPD can provide accurate modelling of water static and dynamic properties and an EOS that matches the experimental data to a large range of confinement pressure.

42 ENGINEERING↗
DOE OSTI2022

Workhorse minimally empirical dispersion-corrected density functional with tests for weakly bound systems: r 2 SCAN + rVV 10

SCAN+rVV10 has been demonstrated to be a versatile van der Waals (vdW) density functional that delivers good predictions of both energetic and structural properties for many types of bonding. Recently, the r 2 SCAN functional was devised as a revised form of SCAN with improved numerical stability. In this work, we refit the rVV10 functional to optimize the r 2 SCAN+rVV10 vdW density functional and test its performance for molecular interactions and layered materials. Our molecular tests demonstrate that r 2 SCAN+rVV10 outperforms its predecessor SCAN+rVV10 in both efficiency (numerical stability) and accuracy. This good performance is also found in lattice-constant predictions. In comparison with benchmark results from higher-level theories or experiments, r 2 SCAN+rVV10 yields excellent interlayer binding energies and phonon dispersions for layered materials.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Transient Stability-Enhancing Method for Grid-Forming Inverters Under Current Limiting

The primary controller of a grid-forming (GFM) inverter governs the internal reference voltage and angle, which enables the inverter to naturally synchronize and share power with the connected grid; however, during disturbances on the grid caused by, e.g., line faults, overloading, or frequency and phase shifts, a GFM inverter can be pushed into a current-limited operation to protect the device against thermal damage. This can lead to a windup of the internal reference frequency generated by the primary controller and cause a total loss of synchronism with the grid. Here, to make the primary controller of GFM inverters more robust against transient instability, this paper proposes a method that manipulates the reference signals governed by the primary controller by introducing the concept of fictitious power. The fictitious power counteracts the integrator windup in the primary controller, which enables the current-limited inverter to conceive a stable operating point. This method is benchmarked against five state-of-the-art GFM inverter stabilization methods to evaluate its stability-enhancing performance. Full-order electromagnetic-transient simulations and experimental results of a hardware inverter test setup validate the benefits of the method to improve the transient stability of GFM inverters in the face of grid disturbances.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2025

Unified differentiable digital twin for the IOTA/FAST facility

As the design complexity of modern accelerators grows, there is more interest in using advanced simulations that have fast execution time or produce insights about accelerator state. One notable example of additional information are gradients of physical observables with respect to design parameters produced by differentiable simulations. The IOTA/FAST facility has recently begun a program to implement and experimentally validate a unified start-to-end differentiable digital twin to serve as a virtual accelerator test stand, allowing for rapid prototyping of new software and experiments with minimal beam time costs. In this contribution we will discuss our plans and progress. Specifically, we will cover the selection and benchmarking of both physics and ML codes, the development of generic interfaces between device models and surrogate or physics-based sections, and the export of the parameters through either a deterministic event loop or a fully asynchronous EPICS soft input/output controller. We will also discuss challenges in model calibration and uncertainty quantification, as well as future plans to support larger proton accelerators like PIPII and Booster.

Kuklev, Nikita [Fermilab]↗
DOE OSTI2021

Investigation of Thermal-Hydraulic Effects of Dry Storage Canister Helium Backfill Loss Using the Horizontal Dry Cask Simulator

A previous investigation produced data sets that can be used to benchmark the codes and best practices presently used to determine cladding temperatures and induced cooling air flows in modern horizontal dry storage systems. The horizontal dry cask simulator (HDCS) was designed to generate this benchmark data and add to the existing knowledge base. The objective of the previous HDCS investigation was to capture the dominant physics of a commercial dry storage system in a well-characterized test apparatus for a wide range of operational parameters. The close coupling between the thermal response of the canister system and the resulting induced cooling air flow rate was of particular importance. The previous investigation explored these parameters using helium backfill at 100 kPa and 800 kPa pressure as well as air backfill with a series of simulated decay heats. The helium tests simulated a horizontal dry cask storage system at normal storage conditions with either atmospheric or elevated backfill pressure, while the air tests simulated horizontal storage canisters following a complete loss of helium backfill, in which case the helium would be replaced by air. The present HDCS investigation adds to the previous investigation by exploring steady-state conditions at various stages of the loss of helium backfill from a horizontal dry cask storage system. This is achieved by using helium/air blends as a backfill in the HDCS and running a series of tests using various simulated decay heats to explore the effects of relative helium/air molar concentration on the thermal response of a simulated horizontal dry cask storage system. A total of twenty tests were conducted where the HDCS achieved steady state for various assembly powers, representative of decay heat. The power levels tested were 0.50, 1.00, 2.50, and 5.00 kW. All tests were run at 100 kPa vessel pressure. The backfill gases used in these tests are given in this report as a function of mole fraction of helium (He), balanced by air: 1.0, 0.9, 0.5, 0.1, and 0.0 He. Steady-state conditions (where the steady-state start condition is defined as where the change in temperature with respect to time for the majority of HDCS components is less than or equal to 0.3 K/h) were achieved for all test cases.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2021

Performance Characteristics of the BlueField-2 SmartNIC

High-performance computing (HPC) researchers have long envisioned scenarios where application workflows could be improved through the use of programmable processing elements embedded in the network fabric. Recently, vendors have introduced programmable Smart Network Interface Cards (SmartNICs) that enable computations to be offloaded to the edge of the network. There is great interest in both the HPC and high-performance data analytics (HPDA) communities in understanding the roles these devices may play in the data paths of upcoming systems. This paper focuses on characterizing both the networking and computing aspects of NVIDIA’s new BlueField-2 SmartNIC when used in a 100Gb/s Ethernet environment. For the networking evaluation we conducted multiple transfer experiments between processors located at the host, the SmartNIC, and a remote host. These tests illuminate how much effort is required to saturate the network and help estimate the processing headroom available on the SmartNIC during transfers. For the computing evaluation we used the stress-ng benchmark to compare the BlueField-2 to other servers and place realistic bounds on the types of offload operations that are appropriate for the hardware. Our findings from this work indicate that while the BlueField-2 provides a flexible means of processing data at the network’s edge, great care must be taken to not overwhelm the hardware. While the host can easily saturate the network link, the SmartNIC’s embedded processors may not have enough computing resources to sustain more than half the expected bandwidth when using kernel-space packet processing. From a computational perspective, encryption operations, memory operations under contention, and on-card IPC operations on the SmartNIC perform significantly better than the general-purpose servers used for comparisons in our experiments. Therefore, applications that mainly focus on these operations may be good candidates for offloading to the SmartNIC.

97 MATHEMATICS AND COMPUTING↗