Search NASA⌕ Search

SEARCH · Search NASA

Results for “computer system benchmarking”

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 577 records · Page 32

Energetically consistent model reduction for metriplectic systems

The metriplectic formalism is useful for describing complete dynamical systems which conserve energy and produce entropy. This creates challenges for model reduction, as the elimination of high-frequency information will generally not preserve the metriplectic structure which governs long-term stability of the system. Based on proper orthogonal decomposition, a provably convergent metriplectic reduced-order model is formulated which is guaranteed to maintain the algebraic structure necessary for energy conservation and entropy formation. Further, numerical results on benchmark problems show that the proposed method is remarkably stable, leading to improved accuracy over long time scales at a moderate increase in cost over naive methods.

42 ENGINEERING↗

Thermal Management for FPGA Nodes in HPC Systems

The integration of FPGAs into large-scale computing systems is gaining attention. In these systems, real-time data handling for networking, tasks for scientific computing, and machine learning can be executed with customized datapaths on reconfigurable fabric within heterogeneous compute nodes. At the same time, thermal management, particularly battling the cooling cost and guaranteeing the reliability, is a continuing concern. The introduction of new heterogeneous components into HPC nodes only adds further complexities to thermal modeling and management. The thermal behavior of multi-FPGA systems deployed within large compute clusters is less explored. Here, we first show that the thermal behaviors of different FPGAs of the same generation can vary due to their physical locations in a rack and process variation, even though they are running the same tasks. We present a machine learning–based model to capture the thermal behavior of each individual FPGA in the cluster. We then propose two thermal management strategies guided by our thermal model. First, we mitigate thermal variation and hotspots across the cluster by proactive thermal-aware task placement. Under the tested system and benchmarks, we achieve up to 26.4° C and on average 13.3° C system temperature reduction with no performance penalty. Second, we utilize this thermal model to guide HLS parameter tuning at the task design stage to achieve improved thermal response after deployment.

97 MATHEMATICS AND COMPUTING↗

One- and two-qubit gate infidelities due to motional errors in trapped ions and electrons

In this work, we derive analytic formulas that determine the effect of error mechanisms on one- and two-qubit gates in trapped ions and electrons. First, we analyze and derive expressions for the effect of driving field inhomogeneities on one-qubit gate fidelities. Second, we derive expressions for two-qubit gate errors, including static motional frequency shifts, trap anharmonicities, field inhomogeneities, heating, and motional dephasing. We show that, for small errors, each of our expressions for infidelity converges to its respective numerical simulation; this shows that our formulas are sufficient for determining error budgets for high-fidelity gates, obviating numerical simulations in future projects. All of the derivations are general to any internal qubit state, and any mixed state of the ion crystal's motion that is diagonal in the Fock state basis. Our treatment of static motional frequency shifts, trap anharmonicities, heating, and motional dephasing apply to both laser-based and laser-free gates, while our treatment of field inhomogeneities applies to laser-free systems.

74 ATOMIC AND MOLECULAR PHYSICS↗

Hybrid Programming for Near-Term Quantum Computing Systems

Recent computations involving quantum processing units (QPUs) have demonstrated a series of challenges inherent to hybrid classical-quantum programming, compilation, execution, and verification and validation. Despite considerable progress, system-level noise, limited low-level instructions sets, remote access models, and an overall lack of portability and classical integration presents near-term programming challenges that must be overcome in order to enable reliable scientific quantum computing and support robust hardware benchmarking. In this work, we draw on our experience in programming QPUs to identify common concerns and challenges, and detail best practices for mitigating these challenges within the current hybrid classical-quantum computing paradigm. Following this discussion, we introduce the XACC quantum compilation and execution framework as a hardware and language independent solution that addresses many of these hybrid programming challenges. XACC supports extensible methodologies for managing a variety of programming, compilation, and execution concerns across the increasingly diverse set of QPUs. We use recent nuclear physics simulations to illustrate how the framework mitigates programming, compilation, and execution challenges and manages the complex workflow present in QPU-enhanced scientific applications. Finally, we codify the resulting hybrid scientific computing workflow in order to identify key areas requiring future improvement.

Mccaskey, Alex↗

ELSI: A unified software interface for Kohn–Sham electronic structure solvers

Solving the electronic structure from a generalized or standard eigenproblem is often the bottleneck in large scale calculations based on Kohn-Sham density-functional theory. This problem must be addressed by essentially all current electronic structure codes, based on similar matrix expressions, and by high-performance computation. We here present a unified software interface, ELSI, to access different strategies that address the Kohn-Sham eigenvalue problem. Currently supported algorithms include the dense generalized eigensolver library ELPA, the orbital minimization method implemented in libOMM, and the pole expansion and selected inversion (PEXSI) approach with lower computational complexity for semilocal density functionals. The ELSI interface aims to simplify the implementation and optimal use of the different strategies, by offering (a) a unified software framework designed for the electronic structure solvers in Kohn-Sham density-functional theory; (b) reasonable default parameters for a chosen solver; (c) automatic conversion between input and internal working matrix formats, and in the future (d) recommendation of the optimal solver depending on the specific problem. As a result, comparative benchmarks are shown for system sizes up to 11,520 atoms (172,800 basis functions) on distributed memory supercomputing architectures.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Nuclear Data and Benchmarking Program: Nuclear Data Performance Assessment for Advanced Reactors

Over the last several Evaluated Nuclear Data File (ENDF)/B releases, many updates have been introduced to the nuclear data libraries that can significantly change computational results. These changes can be particularly important for advanced reactor concepts, which are not as thoroughly investigated as those used in light water reactor (LWR) systems. Therefore, a performance assessment of the ENDF/B libraries using the SCALE 6.2.3 code package was conducted under the auspices of the U.S., Department of Energy Office of Nuclear Energy (DOE-NE). The following systems relevant to the advanced reactor community were chosen as subject for investigation in this report: sodium cooled fast reactor systems, graphite moderated high temperature gas-cooled reactors, and several molten salt reactor models. At first, the similarity of the integral performance of the ENDF/B-VII.1 and ENDF/B-VIII.0 nuclear data libraries caused by compensating errors in important differential data was demonstrated. In calculations for a shipping container of high-assay low enriched uranium, the data of individual isotopes were systematically swapped between the two ENDF/B libraries. An eigenvalue difference of up to 450 pcm was found due to the use of 235 U and 238 U from one library and 1 H and 16 O from the other library. This clearly demonstrates a cross-correlation between reaction data sets of different isotopes within a library that should be reported in the evaluations. For sodium-cooled fast reactor (SFR) fuel assemblies, significant eigenvalue (k inf ) differences (200– 450 pcm) were found between calculations using the 2011 ENDF/B-VII.1 data and 2018 ENDF/B-VIII.0 data. These differences were mainly caused by updates of 238 U and 239 Pu neutron cross sections. Results from the multigroup (MG) calculations further revealed that the group structure of the applied MG library strongly influences the MG bias due to the importance of an appropriate energy resolution of the resonances in higher energy ranges. The application of previously used MG libraries requires new verification for advanced reactor simulation in comparisons with reference continuous-energy calculations. Nuclear data uncertainty analyses of these SFR systems resulted in eigenvalue uncertainties between 1,400 and 1,800 pcm, which are three to four times higher than corresponding uncertainties in light water reactor (LWR) systems. The main contributor to this uncertainty was found to be inelastic scattering on 238 U, which shows an uncertainty of up to 50% in the fast energy range. Other important contributors are the scattering reactions of 56 Fe and 23 Na which have so far not appeared in LWR analysis. Since the largest contribution to the eigenvalue uncertainty of SFR systems is coming from scattering reactions, it is expected that uncertainties in the angular scattering distributions also have a significant impact. Those uncertainties are, however, not available for the majority of nuclides, and the capabilities to determine sensitivities to this data are currently underdeveloped. For two graphite moderated high temperature gas-cooled reactor (HTGR) benchmarks, only the ENDF/BVII. 1 calculations resulted in consistent eigenvalues (k eff ) with the corresponding experimental measurements. Eigenvalue differences of about 1,000 pcm were found between the 2006 ENDF/B-VII.0 data and the 2011 ENDF/B-VII.1 data, primarily due to an updated carbon capture cross section in the thermal energy range in the later ENDF/B release. The ENDF/B-VIII.0 eigenvalue was larger than the ENDF/B-VII.1 eigenvalue by about 300 pcm mainly due to updates in multiple 235 U neutron cross sections, with offsetting updates in the 238 U cross sections. With ENDF/B-VIII.0, graphite can be modeled as perfect crystal or with two different porosities. The choice of the graphite evaluation can have a significant influence of the eigenvalue. For a HTGR benchmark, maximum eigenvalue differences as high as 650 pcm due to different porosities were observed. More detailed studies of the impact of the porosity are necessary while considering that the graphite porosity can vary between the used materials and changes as a function of neutron fluence. xii The uncertainty of the HTGR eigenvalues due to nuclear data uncertainties was found between 500 and 600 pcm, with the top contributor being the neutron multiplicity of 235 U. A gap in the form of missing uncertainties in graphite thermal scattering data, that might have significant impact on HTGR reactors, was identified. This gap is being investigated by the DOE-NE Nuclear Data and Benchmarking Program, as documented in a separate report The fast spectrum molten salt reactor calculations revealed similar comparisons as for the SFR assemblies. The eigenvalue uncertainties were significantly influenced by inelastic scattering on 238 U, while the impact of angular scattering distributions is unknown. For the graphite and zirconium hydride moderated system, eigenvalue uncertainties of up to 700 pcm were observed, requiring in depth analyses of the cross sections and corresponding uncertainties of the included materials. In general, it was observed that the flux spectra of the various molten salt reactor systems show significant differences, even between a fresh state and a depleted state. The choice of the energy group structure of MG calculations is therefore highly relevant; the applicability of previously used MG libraries needs to be verified. Due the unavailability of pin power measurements of advanced reactor systems, pin power calculations with both ENDF/B-VII.1 and VIII.0 data were performed for a light water reactor and compared to corresponding measurements. Both calculations show good agreement with the measurements. A comparison between the ENDF/B-VII.1 and VIII.0 results revealed small differences, mostly in the range of about 0.2%.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

TH-E-BRE-01: A 3D Solver of Linear Boltzmann Transport Equation Based On a New Angular Discretization Method with Positivity for Photon Dose Calculation Benchmarked with Geant4

Purpose: The Linear Boltzmann Transport Equation (LBTE) solved through statistical Monte Carlo (MC) method provides the accurate dose calculation in radiotherapy. This work is to investigate the alternative way for accurately solving LBTE using deterministic numerical method due to its possible advantage in computational speed from MC. Methods: Instead of using traditional spherical harmonics to approximate angular scattering kernel, our deterministic numerical method directly computes angular scattering weights, based on a new angular discretization method that utilizes linear finite element method on the local triangulation of unit angular sphere. As a Result, our angular discretization method has the unique advantage in positivity, i.e., to maintain all scattering weights nonnegative all the time, which is physically correct. Moreover, our method is local in angular space, and therefore handles the anisotropic scattering well, such as the forward-peaking scattering. To be compatible with image-guided radiotherapy, the spatial variables are discretized on the structured grid with the standard diamond scheme. After discretization, the improved sourceiteration method is utilized for solving the linear system without saving the linear system to memory. The accuracy of our 3D solver is validated using analytic solutions and benchmarked with Geant4, a popular MC solver. Results: The differences between Geant4 solutions and our solutions were less than 1.5% for various testing cases that mimic the practical cases. More details are available in the supporting document. Conclusion: We have developed a 3D LBTE solver based on a new angular discretization method that guarantees the positivity of scattering weights for physical correctness, and it has been benchmarked with Geant4 for photon dose calculation.

97 MATHEMATICS AND COMPUTING↗

Extension of the DP-RS-Sim Hybrid Method to Time-Variant Structural Reliability Analysis

This paper presents the extension to time-variant reliability applications of a novel hybrid reliability method, namely the Design Point--Response Surface--Simulation (DP-RS-Sim) method. This method combines in an innovative way the design point search used in First- and Second-Order Reliability Methods (FORM/SORM) with the response surface method and appropriate simulation techniques. The proposed method is illustrated by using as benchmark example a nonlinear inelastic single-degree-of-freedom system subjected to Gaussian white noise base excitation from at rest initial conditions. The system is considered to fail when the roof displacement exceeds a predefined deterministic threshold. Estimates of the time-variant failure probability obtained through the DP-RS-Sim method are compared with the FORM-based mean out-crossing rate computations and crude Monte Carlo simulation results considered as reference solution. It is found that the DP-RS-Sim method can provide accurate time-variant failure probability estimates at low computational cost compared to other existing structural reliability methods and is therefore a promising method for large scale time-variant reliability applications.

58 GEOSCIENCES↗

Data-driven, structure-preserving approximations to entropy-based moment closures for kinetic equations

In this study, we present a data-driven approach for approximating entropy-based closures of moment systems from kinetic equations. The proposed closure learns the entropy function by fitting the map between the moments and the entropy of the moment system, and thus does not depend on the spacetime discretization of the moment system or specific problem configurations such as initial and boundary conditions. With convex and C 2 approximations, this data-driven closure inherits several structural properties from entropy-based closures, such as entropy dissipation, hyperbolicity, and H-Theorem. We construct convex approximations to the Maxwell–Boltzmann entropy using convex splines and neural networks, test them on the plane source benchmark problem for linear transport in slab geometry, and compare the results to the standard, entropy-based systems which solve a convex optimization problem to find the closure. Numerical results indicate that these data-driven closures provide accurate solutions in much less computation time than that required by the optimization routine.

97 MATHEMATICS AND COMPUTING↗

Workshop on Addressing Rigor and Reproducibility in Thermal, Heterogeneous Catalysis

Heterogeneous catalysis has long served as the bedrock of the manufacturing of energy carriers, fuels and chemicals, and various technologies for pollution abatement. The significant complexity and variability spanning the entire breadth of catalyst material properties, synthesis methods, characterization techniques, and evaluation procedures, has focused attention on the need to establish community-accepted best practices for ensuring high-quality, benchmarked, and reproducible data. In addition, increased societal urgency to transition to clean energy and reduce greenhouse gas concentrations has incentivized interdisciplinary, convergent, and translational approaches to catalysis research in recent years. Research engineers and scientists with expertise cutting broadly across materials science, chemical synthesis, interfacial science, spectroscopy, and methods of data science and computational simulation, all bring diverse and important perspectives to catalysis research, but often with little awareness of the complexity of catalytic systems, especially in their working environment. As has already occurred in other scientific fields, there has been growing recognition and consensus in the heterogeneous catalysis research community that mechanisms are needed to improve the rigor and reproducibility (R&R) of experimental measurements, to ensure alignment of the broader research community with a common core of best practices specific to the realization of high-quality catalysis research. Similarly, the field is moving rapidly toward computationally informed and data science-driven catalyst design, but the success of implementing such predictive tools hinges on model training and validation rooted in rigorously obtained and reproducible experimental data that are benchmarked to common specifications. As such, this workshop was convened to prepare a report summarizing best practices for reporting data and performing experiments that researchers can use to benchmark, validate, and reproduce data in specific sub-fields of thermal, heterogeneous catalysis. Additionally, we discussed recommendations for future actions that may improve R&R in this field. The workshop organizers and participants include a diverse range of catalysis researchers from various employment sectors (e.g., academia, industry, national laboratory), institutional mission and resources (e.g., PhD-granting research universities, non-PhD-granting teaching universities), career stage (e.g., early, mid and late-career), technical expertise, and demographic background. This diverse group was involved in the discussion of workshop agenda items, writing this report, and discussing possible future action items for the community to consider, which helped ensure that a broad range of perspectives were captured in the description of the problems at hand and the creation of actionable solutions that may be effectively adopted by the diverse practitioners in catalysis research. Importantly, this group of workshop participants also included very early career researchers (e.g., senior PhD students, postdoctoral scholars) who will become the next generation of scientific leaders in various sectors, thus capturing emerging perspectives of newcomers to the field to shape its future while positively impacting the development of its future workforce. We envision that this effort will help advance the field of catalysis science by improving the rigor and reproducibility of experimental data collected by current researchers and future newcomers to the field, which is of broad importance to health and vitality of any scientific discipline. Therefore, best practices identified in this endeavor for thermal heterogeneous catalysis can be translated to such efforts in other areas of catalysis and other scientific fields involving the study of materials, and vice versa. We also envision this to be an ongoing effort, with future workshops that are convened to discuss issues of rigor and reproducibility on technical topics that were unable to be covered in this workshop due to its scope limitations, and as emerging methods and materials become more prevalent in the research community.

36 MATERIALS SCIENCE↗

LDRD final report : managing shared memory data distribution in hybrid HPC applications.

MPI is the dominant programming model for distributed memory parallel computers, and is often used as the intra-node programming model on multi-core compute nodes. However, application developers are increasingly turning to hybrid models that use threading within a node and MPI between nodes. In contrast to MPI, most current threaded models do not require application developers to deal explicitly with data locality. With increasing core counts and deeper NUMA hierarchies seen in the upcoming LANL/SNL 'Cielo' capability supercomputer, data distribution poses an upper boundary on intra-node scalability within threaded applications. Data locality therefore has to be identified at runtime using static memory allocation policies such as first-touch or next-touch, or specified by the application user at launch time. We evaluate several existing techniques for managing data distribution using micro-benchmarks on an AMD 'Magny-Cours' system with 24 cores among 4 NUMA domains and argue for the adoption of a dynamic runtime system implemented at the kernel level, employing a novel page table replication scheme to gather per-NUMA domain memory access traces.

99 GENERAL AND MISCELLANEOUS↗

DRACS thermal performance evaluation for FHR

Direct Reactor Auxiliary Cooling System (DRACS) is a passive decay heat removal system proposed for the Fluoride-salt-cooled High-temperature Reactor (FHR) that combines coated particle fuel and a graphite moderator with a liquid fluoride salt as the coolant. The DRACS features three coupled natural circulation/convection loops, relying completely on buoyancy as the driving force. These loops are coupled through two heat exchangers, namely, the DRACS Heat Exchanger and the Natural Draft Heat Exchanger. In addition, a fluidic diode is employed to minimize the parasitic flow into the DRACS primary loop and correspondingly the heat loss to the DRACS during normal operation of the reactor, and to keep the DRACS ready for activation, if needed, during accidents. To help with the design and thermal performance evaluation of the DRACS, a computer code using MATLAB has been developed. This code is based on a one-dimensional formulation and its principle is to solve the energy balance and integral momentum equations. By discretizing the DRACS system in the axial direction, a bulk mean temperature is assumed for each mesh cell. The temperatures of all the cells, as well as the mass flow rates in the DRACS loops, are predicted by solving the governing equations that are obtained by integrating the energy conservation equation over each cell and integrating the momentum conservation equation over each of the DRACS loops. In addition, an intermediate heat transfer loop equipped with a pump has also been modeled in the code. This enables the study of flow reversal phenomenon in the DRACS primary loop, associated with the pump trip process. Experimental data from a High-Temperature DRACS Test Facility (HTDF) are not available yet to benchmark the code. A preliminary code validation is performed by using natural circulation experimental data available in the literature, which are as closely relevant as possible. The code is subsequently applied to the HTDF that is under construction at the Ohio State University.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

Protein C-GeM: A Coarse-Grained Electron Model for Fast and Accurate Protein Electrostatics Prediction

The electrostatic potential (ESP) is a powerful property for understanding and predicting electrostatic charge distributions that drive interactions between molecules. In this study, we compare various charge partitioning schemes including fitted charges, density-based quantum mechanical (QM) partitioning schemes, charge equilibration methods, and our recently introduced coarse-grained electron model, C-GeM, to describe the ESP for protein systems. Furthermore, when benchmarked against high quality density functional theory calculations of the ESP for tripeptides and the crambin protein, we find that the C-GeM model is of comparable accuracy to ab initio charge partitioning methods, but with orders of magnitude improvement in computational efficiency since it does not require either the electron density or the electrostatic potential as input.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Overcoming Barriers in Electrochemical Toluene Hydrogenation for Efficient Hydrogen Storage by Pt 3 Au Alloy Catalysts

Hydrogen storage and transportation are essential for the hydrogen economy, and liquid organic hydrogen carriers (LOHCs), such as a toluene/methylcyclohexane (TOL/MCH) system, offer significant advantages in terms of safety and efficiency. However, the electrochemical reduction of TOL to MCH (TER) faces challenges from competing with the hydrogen evolution reaction (HER) and catalyst instability. Here, in this study, Pt 3 Au is introduced as a highly effective catalyst for TER. Through density functional theory screening, we identified distinctive properties of Pt 3 Au, including enhanced binding to the TER intermediates and effective HER suppression. Experimental validation confirmed these computational predictions, with Pt 3 Au achieving the highest reported Faradaic efficiency (98%) in proton exchange membrane systems. Moreover, long-term testing demonstrated that Pt 3 Au maintained Faradaic efficiencies of >90% over 9 h, highlighting its robustness and operational stability. By integrating computational modeling and experimental evaluation, this work addresses key limitations in LOHC catalysis. Pt 3 Au establishes a benchmark for selective and stable TER performance, paving the way for advanced hydrogen storage technologies. These findings emphasize the critical role of rational catalyst design in overcoming the challenges associated with scalable and efficient hydrogen storage solutions.

LOHC↗

Coupled-cluster based approach for core-level states in condensed phase: Theory and application to different protonated forms of aqueous glycine

A theoretical approach for calculating core-level states in condensed phase is presented. The approach is based on equation-of-motion coupled-cluster theory (EOMCC) and effective fragment potential (EFP) method. By introducing an approximate treatment of double excitations in the EOM-CCSD (EOM-CC with single and double substitutions) ansatz, we address poor convergence issues that are encountered for the core-level states and significantly reduce computational costs. While the approximations introduce relatively large errors in the absolute values of transition energies, the errors are systematic. Consequently, chemical shifts, changes in ionization energies relative to reference systems, are reproduced reasonably well. By using different protonation forms of solvated glycine as a benchmark system, we show that our protocol is capable of reproducing the experimental chemical shifts with a quantitative accuracy. The results demonstrate that chemical shifts are very sensitive to the solvent interactions and that explicit treatment of solvent, such as EFP, is essential for achieving quantitative accuracy.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Comprehensive evaluation and parametric sensitivity of interatomic potential models for diffusion kinetics of Cr 2 O 3 in molecular dynamics

While molecular dynamics (MD) has proven to be a promising approach to investigate the diffusion properties, the grand challenge resides in evaluating potential model parameters to accurately replicate experimentally measured properties. The Buckingham potential model with Columbic interaction is widely employed in MD simulations of chromia (Cr 2 O 3 ) systems, as it allows for reasonable computational cost and accuracy. However, considering the well-known limitation of classical potential models in simultaneous reproduction of various physical phenomena, further comprehensive evaluation of the potential is required for calculation of diffusion properties. In this study, we benchmark the performance of three different Buckingham models with the experimental data by calculating structural, thermodynamic, and mechanical properties of defect-free Cr 2 O 3 , and diffusion properties of Cr 2 O 3 with vacancy defects. Available Buckingham models display limited accuracies, consolidating the necessity of retraining the potential parameters for all properties impacting the diffusion dynamics. Oversimplification in parameterization procedures is suggested to impede the universal performance in property reproduction. Furthermore, this research also demonstrates effective guidelines for choosing a proper parameter set of current Buckingham potential for MD simulation with Cr 2 O 3 depending on properties and for potential reparameterization.

36 MATERIALS SCIENCE↗

Early experience on using glidein WMS in the cloud

Cloud computing is steadily gaining traction both in commercial and research worlds, and there seems to be significant potential to the HEP community as well. However, most of the tools used in the HEP community are tailored to the current computing model, which is based on grid computing. One such tool is glideinWMS, a pilot-based workload management system. In this paper we present both what code changes were needed to make it work in the cloud world, as well as what architectural problems we encountered and how we solved them. Benchmarks comparing grid, Magellan, and Amazon EC2 resources are also included.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

An evaluation of the CORAL interconnects

The US Department of Energy deployed the Summit and Sierra supercomputers with the latest state-of-the-art network interconnect technology in 2018 and both systems entered production in 2019. In this paper, we provide an in-depth assessment of the systems' network interconnects that are based on Enhanced Data Rate (EDR) 100 Gb/s Mellanox InfiniBand. Both systems use second-generation EDR Host Channel Adapters (HCAs) and switches with several new features such as Adaptive Routing (AR), switch-based collectives, and HCA-based tag matching. Although based on the same components, Summit's network is "non-blocking" (i.e., a fully provisioned Clos network) and Sierra's network has a 2:1 taper between the racks and aggregation switches. We evaluate the two systems' interconnects using traditional communication benchmarks as well as production applications. We find that the new Adaptive Routing dramatically improves performance but the other new features still need improvement.

Zimmer, Christopher↗