Search NASA⌕ Search

SEARCH · Search NASA

Results for “High Performance Storage Systems”

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 703 records · Page 39

Development of High Capacity Energy Storage Materials

Hydrogen fuel cells have the potential to offer energy and power density advantages over lithium ion batteries in automotive and portable power applications when paired with an appropriate hydrogen storage system. Development of the ideal hydrogen storage material has been immensely sought after but plagued by limitations present in each type of material. These limitations typically include nonidealistic operational temperatures, low capacities, excessive costs, lack of reversibility, or evolution of impurities which irreversibly damage fuel cell performance. Many complex metal hydrides possess suitable hydrogen capacities but unfortunately suffer from impurity release. In order to address this problem, our team has explored in-situ catalyst doping of alkali metal amides to demonstrate methodologies to reduce ammonia release. The incorporation of iridium metal in lithium amide significantly reduces the ammonia release and decomposes the ammonia to provide additional hydrogen. Furthermore, we demonstrate, for the first time, high resistance of low temperature fuel cells to ammonia in comparison with typical proton exchange membrane fuel cells. The utilization of these two approaches in tandem provides a novel pathway for the development and implementation of high capacity energy storage materials for fuel cell applications.

25 ENERGY STORAGE↗

Recent trends in all-organic polymer dielectrics for high-temperature electrostatic energy storage capacitors

Electrostatic energy storage (EES) capacitors are critical for renewable energy and high-power systems, driving the search for dielectric materials that combine superior electrical insulation, mechanical flexibility, low density, cost-effectiveness, and processability. Polymer-based dielectrics have emerged as leading candidates, particularly for high electric field applications. However, conventional polymers often fail to meet the demands of high-temperature environments due to increased electrical conductivity and reduced discharged energy density at elevated temperatures, resulting in energy loss and reduced performance. High glass transition temperature (T g) polymers show promise but require further optimization to enhance their energy storage capabilities under thermal and electrical stress. This review provides a comprehensive update on recent advancements in high-T g polymer-based dielectrics for EES capacitors, focusing on both intrinsic polymers and all-organic composites. It outlines key design principles, critical performance parameters, and innovative strategies—such as nanofiller doping, layered architectures, physical blending, and chemical crosslinking—to improve electrical, thermal, and mechanical properties. The review also highlights emerging trends, including the integration of machine learning algorithms to explore novel polymer structures and expand the chemical design space. By bridging the gap between academic research and industrial application, this review aims to accelerate the development of next-generation dielectric materials capable of balancing multiple performance metrics for high-temperature EES capacitors.

Xie, Zongliang↗

A novel battery scheme: Coupling nanostructured phosphorus anodes with lithium sulfide cathodes

Lithium-ion batteries are approaching their theoretical limit and can no longer keep up with the increasing demands of human society. Lithium-sulfur batteries, with a high theoretical specific energy, are promising candidates for next generation energy storage. However, the use of Li metal in Li-S batteries compromises both safety and performance, enabling dendrite formation and causing fast capacity degradation. Previous studies have probed alternative battery systems to replace the metallic Li in Li-S system, such as a Si/Li2S couple, with limited success in performance. Recently, there is a focus on red P as a favorable anode material to host Li. In this work, we establish a novel battery scheme by utilizing a P/C nanocomposite anode and pairing it with a Li 2 S coated carbon nanofiber cathode. We find that red P anode can be compatible in ether-based electrolyte systems and can be successfully coupled to a Li 2 S cathode. Our proof of concept full-cell displays remarkable specific capacity, rate and cycling performances. We expect our work will provide a useful alternative system and valuable insight in the quest for next generation energy storage devices.

25 ENERGY STORAGE↗

Membrane design for non-aqueous redox flow batteries: Current status and path forward

Redox flow batteries are promising technologies for large-scale, long-duration energy storage applications. Among them, non-aqueous redox flow batteries (NARFB) represent a transformative flow battery system since NARFBs potentially offer a higher energy density than aqueous flow batteries. However, many technical challenges remain for NARFBs, including the lack of high-performance membranes, low solubility of redox materials, and poor cycling efficiencies. Membranes serve a vital function in NARFBs, as they allow for selective ion transport while providing separation between the anolyte and catholyte. NARFB membrane development is an emerging research area, and this article reviews their design and critical factors that influence membrane properties, including solvent uptake, ion transport, and redox species permeability. A greater understanding of membrane behavior in non-aqueous solutions provides design principles for developing next-generation membranes for NARFB. In conclusion, we summarize the challenges, target metrics, and future perspectives for NARFBs.

38 RADIATION CHEMISTRY, RADIOCHEMISTRY, AND NUCLEA↗

Pore-scale investigation on stress-dependent characteristics of granular packs and the impact of pore deformation on fluid distribution

Understanding the effect of changing stress conditions on multiphase flow in porous media is of fundamental importance for many subsurface activities including enhanced oil recovery, water drawdown from aquifers, soil confinement, and geologic carbon storage. Geomechanical properties of complex porous systems are dynamically linked to flow conditions, but their feedback relationship is often oversimplified due to the difficulty of representing pore-scale stress deformation and multiphase flow characteristics in high fidelity. In this work, we performed pore-scale experiments of single- and multiphase flow through bead packs at different confining pressure conditions to elucidate compaction-dependent characteristics of granular packs and their impact on fluid flow. A series of drainage and imbibition cycles were conducted on a water-wet, soda-lime glass bead pack under varying confining stress conditions. Simultaneously, X-ray micro-CT was used to visualize and quantify the degree of deformation and fluid distribution corresponding with each stress condition and injection cycle. Micro-CT images were segmented using a gradient-based method to identify fluids (e.g., oil and water), and solid phase redistribution throughout the different experimental stages. Changes in porosity, tortuosity, and specific surface area were quantified as a function of applied confining pressure. Results demonstrate varying degrees of sensitivity of these properties to confining pressure, which suggests that caution must be taken when considering scalability of these properties for practical modeling purposes. Changes in capillary number with confining pressure are attributed to the increase in pore velocity as a result of pore contraction. Furthermore, this increase in pore velocity was found to have a marginal impact on average phase trapping at different confining pressures.

58 GEOSCIENCES↗

Vehicle Technologies Program: Energy Storage R&D (2008 Annual Progress Report)

One of the primary objectives of the Energy Storage effort is the development of durable and affordable advanced batteries (and ultracapacitors) for use in a full range of vehicle applications, from start/stop to full-power HEVs, EVs, and PHEVs. The battery technology development activity spans three areas: system development of full battery systems; benchmark testing of emerging technologies in order to remain abreast of the latest industry developments; and Small Business Innovative Research (SBIR) to fund early-stage R&D for small businesses/entrepreneurs.

25 ENERGY STORAGE↗

HERMES: A cyber-energy SAAS platform to provide real-time 2d and 3d interactive visualizations [SWR-23-15]

HERMES is a novel visualization SAAS platform for achieving real-time visualization of large-scale environments involving cyber-energy devices. The platform receives data from emulated environments which include physical hardware and network devices that represent a complex IT / OT system. The platform is capable of streaming, filtering, storing, and visualizing all data within the environment and scaling vertically and horizontally. This capability enables high fidelity visual analysis of events to be performed in real time as well as the collection and storage of historical data for forensic analysis.

Van Natta, Joshua↗

Integrated End-to-end Performance Prediction and Diagnosis for Extreme Scientific Workflows

This report details recent progress for the ASCR funded project “Integrated End-to-end Performance Prediction and Diagnosis for Extreme Scientific Workflows”. We refer to the project as IPPD/2, reflecting the 2017 renewal under expanded scope and partners In IPPD/2, we increased our research scope to include data motion. We are focusing on three major aspects: a) observe how data is generated, distributed, and used; b) analyze how data is (repeatedly) consumed with a focus both on repeated patterns and anomalies; and c) explore how to optimize data motion. This new work on data motion will augment and complement IPPD/2’s research that focused on the computational aspects of tasks. We leverage and extend our existing tools and demonstrate our work on the Belle II workflow suite as well as on workflows from NSLS-II. The highlights of our work are as follows: Provenance for Workflows: Provenance is used to provide information enabling quality control, re-run computational workflows, and reproduce results. IPPD/2 has been building a scalable provenance management system that enables the capture of provenance from the high-level workflow through all relevant system levels in one integrated environment. Leveraging this work, our recent efforts have included using provenance as an enabling technique. Workload characterization: Leveraging provenance and analysis, we characterize data movement within network, storage, and memory over a variety of workloads. This characterization enables an understanding by performance analysts and application developers of the range of behaviors that could be expected. Performance Prediction for Workflows: The goal of modeling distributed workflows is to understand performance bottlenecks and enable more intelligent task scheduling to optimize selected metrics of interest (e.g., task throughput or output data rate). IPPD/2 has utilized both analytical and AI/ML modeling methodologies for performance modeling. Advanced Scheduling and Fault Modeling for Workflows: Scheduling of large-scale scientific workflows on geographically distributed resources is a challenging problem. To improve workflow throughput, we combined novel scheduling algorithms with task predictions from performance modeling and fault modeling. Dynamically Alleviating Bottlenecks in Workflows: Exploiting our provenance, analysis, and modeling efforts, we have explored and developed several techniques for dynamically detecting and alleviating bottlenecks in data movement. In particular, we have spent considerable effort demonstrating our techniques on production-like workflow configurations.

97 MATHEMATICS AND COMPUTING↗

Battle of the Defaults: Extracting Performance Characteristics of HDF5 under Production Load

Popular parallel I/O libraries, such as HDF5, provide tuning parameters to obtain superior performance. However, the selection of effective parameters on production systems is complex due to the interdependence of I/O software and file system layers. Hence, application developers typically use the default parameters and often experience poor I/O performance. This work conducts a benchmarking-based analysis on the HDF5 behaviors with a wide variety of I/O patterns to extract performance characteristics under the production workload. To make the analysis well controlled, we exercise I/O benchmarks on POSIX-IO, MPI-IO, and HDF5 using the same I/O patterns and in the same jobs. To address high performance variability in production environments, we repeat the benchmarks across I/O patterns, storage devices, and time intervals. Based on the results, we identified consistent HDF5 behaviors that appropriate configurations and operations on dataset layout and file-metadata placement can improve performance significantly. We apply our findings and evaluate the tuned I/O library on two supercomputers: Summit and Cori. The results show that our tuned parameters can achieve more than 10× I/O performance speedup than that with default parameters on both systems, suggesting the effectiveness, stability, and generality of our solution.

Xie, Bing↗

Battery Lifetime Analysis and Simulation Tool (BLAST) Documentation

The deployment and use of lithium-ion (Li-ion) batteries in automotive and stationary energy storage applications must be optimized to justify their high up-front costs. Given that batteries degrade with use and storage, such optimizations must evaluate many years of operation. As the degradation mechanisms are sensitive to temperature, state-of-charge (SOC) histories, current levels, and cycle depth and frequency, it is important to model both the battery and the application to a high level of detail to ensure battery response is accurately predicted. To address these issues, the National Renewable Energy Laboratory (NREL) has developed the Battery Lifetime Analysis and Simulation Tool (BLAST) suite. This suite of tools pairs NREL’s high-fidelity battery degradation model with a battery electrical and thermal performance model, application-specific electrical and thermal performance models of the larger system (e.g., an electric vehicle), application-specific system use data (e.g., vehicle travel patterns and driving data), and historic climate data from cities across the United States. This provides highly realistic long-term predictions of battery response and thereby enables quantitative comparisons of varied battery use strategies.

25 ENERGY STORAGE↗

SU-E-T-152: Building An Ultrafast, Earthquake Proof Treatment Planning System

Purpose: Situated 20 miles from 5 major fault lines in California’s Bay Area, Stanford Universityhas a critical need for IT infrastructure planning to handle the high probability devastating earthquakes. Recently, a multi-million dollar project has been underway to overhaul Stanford’s radiation oncology information systems, maximizing planning system performance and providing true disaster recovery abilities. An overview of the project will be given with particular focus on lessons learned throughout the build. Methods: In this implementation, two isolated external datacenters provide geographical redundancy to Stanford’s main campus datacenter. Real-time mirroring is made of all data stored to our serial attached network (SAN) storage. In each datacenter, hardware/software virtualization was heavily implemented to maximize server efficiency and provide a robust mechanism to seamlessly migrate users in the event of an earthquake. System performance is routinely assessed through the use of virtualized data robots, able to log in to the system at scheduled times, perform routine planning tasks and report timing results to a performance dashboard. A substantial dose calculation framework (608 CPU cores) has been constructed as part of the implementation. Results: Migration to a virtualized server environment with a high performance SAN has resulted in up to a 45% speed up of common treatment planning tasks. Switching to a 608 core DCF has resulted in a 280% speed increase in dose calculations. Server tuning was found to further improved read/write performance by 20%. Disaster recovery tests are carried out quarterly and, although successful, remain time consuming to perform and verify functionality. Conclusion: Achieving true disaster recovery capabilities is possible through server virtualization, support from skilled IT staff and leadership. Substantial performance improvements are also achievable through careful tuning of server resources and disk read/write operations. Developing a streamlined method to comprehensively test failover is a key requirement to the system’s success.

60 APPLIED LIFE SCIENCES↗

FePt-based exchange-coupled composite perpendicular recording media.

In order to demonstrate the strong and proper exchange coupling between bottom magnetic layer and capping magnetic layer in FePt-based exchange-coupled composite (ECC) perpendicular recording media, we have investigated the thermal stability, writability, degree of exchange coupling, and ECC gain factor of FePt-based ECC media as a function of the thickness of the capping FePt layer. With increasing the thickness of capping FePt layer, both thermal stability and media writability increase simultaneously as a result of the change from strong exchange coupling to proper exchange coupling. The proper exchange-coupled FePt ECC is a promising candidate for future high-density perpendicular recording.

36 MATERIALS SCIENCE↗

...And Eat it Too: High Read Performance in Write-Optimized HPC I/O Middleware File Formats

As HPC applications run on increasingly high process counts on larger and larger machines, both the frequency of checkpoints needed for fault tolerance and the resolution and size of Data Analysis Dumps are expected to increase proportionally. In order to maintain an acceptable ratio of time spent performing useful computation work to time spent performing I/O, write bandwidth to the underlying storage system must increase proportionally to this increase in the checkpoint and computation size. Unfortunately, popular scientific self-describing file formats such as netCDF and HDF5 are designed with a focus on portability and flexibility. Extra care and careful crafting of the output structure and API calls is required to optimize for write performance using these APIs. To provide sufficient write bandwidth to continue to support the demands of scientific applications, the HPC community has developed a number of I/O middleware layers, that structure output into write-optimized file formats. However, the obvious concern with any write optimized file format would be a corresponding penalty on reads. In the log-structured filesystem, for example, a file generated by random writes could be written efficiently, but reading the file back sequentially later would result in very poor performance. Simulation results require efficient read-back for visualization and analytics, and though most checkpoint files are never used, the efficiency of a restart is very important in the face of inevitable failures. The utility of write speed improving middleware would be greatly diminished if it sacrificed acceptable read performance. In this paper we examine the read performance of two write-optimized middleware layers on large parallel machines and compare it to reading data natively in popular file formats. The two I/O middleware layer examined in this paper are the Adaptable IO System (ADIOS), a library-based approach developed at Oak Ridge National Laboratory to provide a high-level IO API that can be used in place of netCDF or HDF5 to do much more aggressive write-behind and efficient reordering of data locations within the file; and the Parallel Log-structured Filesystem (PLFS), a stackable FUSE filesystem approach developed at Los Alamos National Laboratory that decouples concurrent writes to improve the speed of checkpoints. Since ADIOS is an I/O componentization that affords selection of different I/O methods at or during runtime, through a single API, users can have access to MPI-IO, Posix-IO, HDF5, netCDF, and staging methods. The ADIOS BP file format is a new log-file format that has a superset of the features of both HDF5 and netCDF, but is designed to be portable and flexible while being optimized for writing. PLFS, on the other hand, is mounted as a stackable filesystem on top of an existing parallel filesystem. Reads or writes to the PLFS filesystem are transparently translated into operations on per-process log files stored in the underlying parallel filesystem. Since PLFS performs this translation without application modification, users can write in HDF5, netCDF, or app-specific file formats and PLFS will store the writes in a set of efficiently written log-formatted files, while presenting the user with a logical 'flat' file on reads. Despite their different approaches, the commonality behind both of these middleware systems is that they both write to a log file format. As shown in previous publications, writes are fully optimized in both systems, sometimes resulting in 100x improvements over writing data in popular file formats. But as mentioned above, writing is only one part of the story. In this paper we examine the read performance of our middleware layers on large parallel machines and compare these to reading data either natively or from other popular file formats. We compare the reading performance in two different scenarios: (1) Reading back restarts from the same number of processors as wrote the data and (2) Reading back restart data from a different number of processors. We observe that not only can write-optimized I/O middleware be built to not penalize read speeds, but for important workloads, techniques that improve write performance can, perhaps counterintuitively, improve read speeds over reading from a contiguously organized file format. In the remainder of this paper, we investigate this further through case studies of PLFS and ADIOS on simulation checkpoint restarts.

99 GENERAL AND MISCELLANEOUS↗

Analysis of a Fluidized-Bed Particle/Supercritical-CO2 Heat Exchanger in a Concentrating Solar Power System

Concentrating solar power (CSP) development has focused on increasing the energy conversion efficiency and lowering the capital cost. To improve performance, CSP research is moving to high-temperature and high-efficiency designs. One technology approach is to use inexpensive, high-temperature heat transfer fluids and storage, integrated with a high-efficiency power cycle such as the supercritical carbon dioxide (sCO 2 ) Brayton power cycle. The sCO 2 Brayton power cycle has strong potential to achieve performance targets of 50% thermal-to-electric efficiency and dry cooling at an ambient temperature of up to 40 °C and to reduce the cost of power generation. Solid particles have been proposed as a possible high-temperature heat transfer or storage medium that is inexpensive and stable at high temperatures above 1000 °C. The particle/sCO 2 heat exchanger (HX) provides a connection between the particles and sCO 2 fluid in emerging sCO 2 power cycles. This article presents heat transfer modeling to analyze the particle/sCO 2 HX design and assess design tradeoffs including the HX cost. The heat transfer process was modeled based on a particle/sCO 2 counterflow configuration, and empirical heat transfer correlations for the fluidized bed and sCO 2 were used to calculate heat transfer area and estimate the HX cost. A computational fluid dynamics simulation was applied to characterize particle distribution and fluidization. This article shows a path to achieve the cost and performance objectives for a particle/sCO 2 HX design by using fluidized-bed technology.

14 SOLAR ENERGY↗

Insights into the Assembly and Conformation of Nanoparticle Organic Hybrid Materials (NOHMs) in Solution with Varying Grafting Type

Recently, nanoparticle organic hybrid materials (NOHMs) have been considered promising electrolytes for energy storage applications due to their unique combination of properties, which include high thermal stability, negligible vapor pressure, and easy synthesis. However, the structural features of NOHMs in electrochemical systems are not well understood. Hence, we investigate the impact of grafting type (ionic vs covalent) on the structure and performance of NOHMs that consist of a silica core and grafted HPE polymer (Jeffamine M2070) in aqueous solution with and without the presence of the supporting electrolyte (0.1 M KHCO 3 ) using small-angle neutron scattering. Careful analyses of the results indicate that even at low grafting density (0.8 chains nm –2 ) and concentration (1–3 wt %), ionic NOHMs solutions contain a free polymer that may interact with the grafted polymer, altering the assembly of NOHMs in solution. Our investigation also elucidates that the covalent NOHM solutions consist of both aggregated and dispersed NOHMs in the solution. Moreover, the addition of the supporting electrolyte affects the assembly and structure of the polymer in the ionic NOHMs solution significantly, whereas only a slight change in the conformation of the grafted polymer is observed for covalent NOHMs. These conformational changes alter the performance of the NOHMs solution, resulting in a dramatic change in viscosity. Careful analysis shows that this significant alteration of solution viscosity in the ionic NOHMs can be traced back to the presence of the free polymer that interacts with the grafted layer and the attenuation of this interaction between the free and grafted polymers with the addition of salt. Furthermore, these studies elucidate the impact of grafting type on the overall structure, assembly, and performance of functionalized NOHMs in aqueous solution and explicitly correlate the structure and assembly of NOHMs to their transport properties and must be considered when designing functional systems that incorporate NOHMs.

25 ENERGY STORAGE↗

Development of Production-Intent Plug-In Hybrid Vehicle Using Advanced Lithium-Ion Battery Packs with Deployment to a Demonstration Fleet

The primary goal of this project was to speed the development of one of the first commercially available, OEM-produced plug-in hybrid electric vehicles (PHEV). The performance of the PHEV was expected to double the fuel economy of the conventional hybrid version. This vehicle program incorporated a number of advanced technologies, including advanced lithium-ion battery packs and an E85-capable flex-fuel engine. The project developed, fully integrated, and validated plug-in specific systems and controls by using GM’s Global Vehicle Development Process (GVDP) for production vehicles. Engineering Development related activities included the build of mule vehicles and integration vehicles for Phases I & II of the project. Performance data for these vehicles was shared with the U.S. Department of Energy (DOE). The deployment of many of these vehicles was restricted to internal use at GM sites or restricted to assigned GM drivers. Phase III of the project captured the first half or Alpha phase of the Engineering tasks for the development of a new thermal management design for a second generation battery module. The project spanned five years. It included six on-site technical reviews with representatives from the DOE. One unique aspect of the GM/DOE collaborative project was the involvement of the DOE throughout the OEM vehicle development process. The DOE gained an understanding of how an OEM develops vehicle efficiency and FE performance, while balancing many other vehicle performance attributes to provide customers well balanced and fuel efficient vehicles that are exciting to drive. Many vehicle content and performance trade-offs were encountered throughout the vehicle development process to achieve product cost and performance targets for both the OEM and end customer. The project team completed two sets of PHEV development vehicles with fully integrated PHEV systems. Over 50 development vehicles were built and operated for over 180,000 development miles. The team also completed four GM engineering development Buy-Off rides/milestones. The project included numerous engineering vehicle and systems development trips including extreme hot, cold and altitude exposure. The final fuel economy performance demonstrated met the objectives of the PHEV collaborative GM/DOE project. Charge depletion fuel economy of twice that of the non-PHEV model was demonstrated. The project team also designed, developed and tested a high voltage battery module concept that appears to be feasible from a manufacturability, cost and performance standpoint. The project provided important product development and knowledge as well as technological learnings and advancements that include multiple U.S. patent applications.

25 ENERGY STORAGE↗

Influence of Metal Identity on Light-Induced Switchable Adsorption in Azobenzene-Based Metal–Organic Frameworks

Energy-efficient capture and release of small gas molecules, particularly carbon dioxide (CO 2 ) and methane (CH 4 ), are of significant interest in academia and industry. Porous materials such as metal–organic frameworks (MOFs) have been extensively studied, as their ultrahigh porosities and tunability enable significant amounts of gas to be adsorbed while also allowing specific applications to be targeted. However, because of the microporous nature of MOFs, the gas adsorption performance is dominated by high uptake capacity at low pressures, limiting their application. Hence, methods involving stimuli-responsive materials, particularly light-induced switchable adsorption (LISA), offer a unique alternative to thermal methods. Here, we report the mechanism of a well-known LISA system, the azobenzene-based material PCN-250, for CO 2 and CH 4 adsorption. There is a noticeable difference in the LISA effect dependent on the metal cluster involved, with the most significant being PCN-250-Al, where the adsorption can change by 83.1% CH 4 and 56.1% CO 2 at 298 K and 1 bar and inducing volumetric storage changes of 36.2 and 33.9 cm 3 /cm 3 at 298 K between 5 and 85 bar (CH 4 ) and 2 and 9 bar (CO 2 ), respectively. Using UV light in both single-crystal X-ray diffraction and gas adsorption testing, we show that upon photoirradiation, the framework undergoes a “localized heating” phenomenon comparable to an increase of 130 K for PCN-250-Fe and improves the working capacity. Furthermore, this process functions because of the constrained nature of the ligand, preventing the typical trans-to-cis isomerization observed in free azobenzene. In addition, we observed that the degree of localized heating is highly dependent on the metal cluster involved, with the series of isostructural PCN-250 systems showing variable performance based upon the degree of interaction between the ligand and the metal center.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Hvac: Removing I/O Bottleneck for Large-Scale Deep Learning Applications

Scientific communities are increasingly adopting deep learning (DL) models in their applications to accelerate scientific discovery processes. However, with rapid growth in the computing capabilities of HPC supercomputers, large-scale DL applications have to spend a significant portion of training time performing I/O to a parallel storage system. Previous research works have investigated optimization techniques such as prefetching and caching. Unfortunately, there exist non-trivial challenges to adopting the existing solutions on HPC supercomputers for large-scale DL training applications, which include non-performance and/or failures at extreme scale, lack of portability and generality in design, complex deployment methodology, and being limited to a specific application or dataset. To address these challenges, we propose High-Velocity AI Cache (HVAC), a distributed read-cache layer that targets and fully exploits the node-local storage or near node-local storage technology. HVAC seamlessly accelerates read I/O by aggregating node-local or near node-local storage, avoiding metadata lookups and file locking while preserving portability in the application code. We deploy and evaluate HVAC on 1,024 nodes (with over 6000 NVIDIA V100 GPUS) of the Summit supercomputer. In particular, we evaluate the scalability, efficiency, accuracy, and load distribution of HVAC compared to GPFS and XFS-on-NVMe. With four different DL applications, we observe an average 25 % performance improvement atop GPFS and 9% drop against XFS-on-NVMe, which scale linearly and are considered the performance upper bound. We envision HVAC as an important caching library for upcoming HPC supercomputers such as Frontier.

Khan, Awais↗