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 361 records · Page 20

Integrated Systems-to-Atoms (S2A) Framework for Designing Resilient and Efficient Hydrogen Infrastructure Solutions

The success of a future clean hydrogen infrastructure will depend on technology performance, operating conditions, and system configuration, which will be integrated to meet specific end-use requirements. We introduce a Systems-to-Atoms framework to cosimulate material, device, and system design for clean hydrogen storage and transport. Here we present a demonstration scenario in which hydrogen is distributed using a liquid organic hydrogen carrier to 700 bar refueling stations. Cross-scale analysis on hydrogen release at the refueling station reveals that while a high reaction pressure (up to 100 bar) may not enhance the catalyst performance, it can lead to system-level cost savings (up to 5% of total refueling station cost) due to savings in downstream gas compression. Further, a typically low-performing dehydrogenation catalyst material (e.g., copper) may be preferred over a high-performing catalyst (e.g., palladium) at different operating conditions when considering overall system affordability and supply chain resilience.

08 HYDROGEN↗

Thermal and Electrical Properties of Isocyanate Derived Organic Aerogels for Cryogenic Insulation Applications

The use and storage of cryogens such as liquefied nitrogen, helium, hydrogen among others requires reliable and efficient thermal insulation systems. Passive insulation from high performance materials that are well-known for their inherent low thermal conductivity would reduce the overall costs involved in design, manufacture and maintenance of such systems. One such class of materials are referred to as aerogels. These materials are known for their low density, high mesoporosity, high surface areas, low thermal conductivity and high acoustic impedance. Aerogels were invented by S.S. Kistler in 1931 and the most common type are those made of silica. However, the inherent fragility of silica aerogels makes them hard to mass produce, and therefore applications have been limited. A major breakthrough was introduced by our team almost 20 years ago with the invention of polymer crosslinked silica aerogels. Those materials shifted attention to all-polymer aerogels that have overcome all fragility issues associated with their inorganic counterparts. This study focuses on such polymeric aerogels that can be mass produced as large monoliths while maintaining the low thermal conductivity of traditional silica aerogels over a wide temperature range. Manufacturing flexibility of polymeric aerogels allows fabrication of blocks and sheets that can be applied in various configurations to insulate cryogenic and superconducting devices. The thermal conductivity with 80 K and room temperature boundary are reported as well as other properties (electrical), that need to be considered when designing devices for cryogenic applications.

Churu, Gitogo↗

Architecture-Based Control of Temperature Gradient-Driven Water Transport in Polymer Electrolyte Fuel Cells

Temperature gradients can move water within the porous media of a polymer electrolyte fuel cell through phase-change-induced (PCI) flow. It is critical to understand PCI flow so that control and distribution of water within a fuel cell can be accomplished. This work investigates the role of architecture, specifically anode land width, on overall cell water content and distribution in the through-plane direction. A specially-designed 4.8 cm 2 fuel cell with precise thermal boundary conditions was imaged with neutrons using two different anode flow field configurations. A new, non-dimensional thermal transport number was developed which quantifies the relative influence of PCI flow on cell water transport. It was found that anode lands larger than the cathode lands cause large thermal gradients that instigate net water flux from the cathode to the anode. An asymmetric configuration with larger anode lands was found to have large changes in water content that were strongly sensitive to cell operating conditions. The thermal transport number developed here enables deduction of the net flux condition based on operating conditions and architecture. This approach enables design of high-performance systems with balanced water management.

25 ENERGY STORAGE↗

User-level File Systems Specialized for HPC Workloads Year-End Report

High-performance computing (HPC) clusters attract the attention of Deep Learning (DL) training users due to the clusters' powerful computation capabilities. While there have been many existing efforts to enable deep neural works to leverage the powerful CPU and GPU processors from leadership high-performance computing (HPC) systems, large-scale deep learning with larger datasets requires efficient I/O support from the underlying file and storage systems. As some current and upcoming HPC clusters have large on-node memory or are equipped with NVMe SSD on compute nodes, more distributed DL trainings are considered to leverage those storages to store dataset for efficient dataset access. Our project goal is to design a specialized DL-oriented file system to improve datasets loading performance of any DL training applications. Over the past year, the team at Florida State University has performed research activities in three different aspects, including 1) completing a specialized memory-based I/O framework (DeepIO) for improving dataset loading performance on DL applications; 2) proposing a more generalized file system (DLFS) for DL applications on node-local SSDs; 3) completing the BeeGFS performance evaluation project for Deep Neural Networks.

97 MATHEMATICS AND COMPUTING↗

THE NEED FOR A NEW JOINING TECHNOLOGY FOR THE CLOSURE WELDING OF RADIOACTIVE MATERIALS CONTAINERS

One of the activities associated with cleanup throughout the Department of Energy (DOE) complex is packaging radioactive materials into storage containers. Much of this work will be performed in high-radiation environments requiring fully remote operations, for which existing, proven systems do not currently exist. These conditions demand a process that is capable of producing acceptable (defect-free) welds on a consistent basis; the need to perform weld repair, under fully-remote operations, can be extremely costly and time consuming. Current closure welding technology (fusion welding) is not well suited for this application and will present risk to cleanup cost and schedule. To address this risk, Fluor and the Pacific Northwest National Laboratory (PNNL), are proposing that a new and emerging joining technology, Friction Stir Welding (FSW), be considered for this work. FSW technology has been demonstrated in other industries (aerospace and marine) to produce near flaw-free welds on a consistent basis. FSW is judged capable of providing the needed performance for fully-remote closure welding of containers for radioactive materials for the following reasons: FSW is a solid-state process; material is not melted. As such, FSW does not produce the type of defects associated with fusion welding, e.g., solidification-induced porosity, cracking, distortion due to weld shrinkage, and residual stress. In addition, because FSW is a low-heat input process, material properties (mechanical, corrosion and environmental) are preserved and not degraded as can occur with 'high-heat' fusion welding processes. When compared to fusion processes, FSW produces extremely high weld quality. FSW is performed using machine-tool technology. The equipment is simple and robust and well-suited for high radiation, fully-remote operations compared to the relatively complex equipment associated with the fusion-welding processes. Additionally, for standard wall thicknesses of radioactive materials containers, the FSW process can perform final closure welding in a single pass (GTAW requires multiple passes) resulting in increased productivity. Together, the performance characteristics associated with FSW, i.e., high weld quality, simple machine-tool equipment and increased welding efficiency, are expected to reduce risk to upcoming DOE radioactive materials packaging campaigns. FSW technology requires some development/adaptation for this application, along with approval from the governing code of construction prior to production operations. This paper addresses the need for a new joining technology, a description of the FSW process and why it is well-suited for this application, and several activities required for commercialization.

42 ENGINEERING↗

CODE ACCEPTANCE OF A NEW JOINING TECHNOLOGY FOR STORAGE CONTAINMENTS [REISSUE]

One of the activities associated with cleanup throughout the Department of Energy (DOE) complex is packaging radioactive materials into storage containers. Much of this work will be performed in high-radiation environments requiring fully remote operations, for which existing, proven systems do not currently exist. These conditions require a process that is capable of producing acceptable (defect-free) welds on a consistent basis; the need to perform weld repair, under fully-remote operations, can be extremely costly and time consuming. Current closure-welding technologies (fusion welding) are not well suited for this application and will present risk to cleanup cost and schedule. To address this risk, Fluor and the Pacific Northwest National Laboratory (PNNL) are proposing that a new and emerging joining technology, Friction Stir Welding (FSW), be considered for this work. FSW technology has been demonstrated in other industries (aerospace and marine) to produce near flaw-free welds on a consistent basis. FSW is judged capable of providing the needed performance for fully-remote closure welding of containers for radioactive materials for the following reasons: FSW is a solid-state process; material is not melted. FSW does not produce the type of defects associated with fusion welding, e.g., solidification-induced porosity, cracking, and distortion due to weld shrinkage. In addition, because FSW is a low-heat input process, material properties (mechanical, corrosion and environmental) experience less degradation in the heat affected zones than do fusion welds. When compared to fusion processes, FSW produces extremely high weld quality. FSW is performed using machine-tool technology. The equipment is simple and robust and well-suited for high radiation, fully-remote operations compared to the relatively complex equipment associated with fusion-welding processes. Additionally, for standard wall thicknesses of radioactive materials containers, the FSW process can perform the final closure in a single pass (GTAW requires multiple passes) resulting in increased productivity. The performance characteristics of FSW, Le., high weld quality, simple machine-tool equipment and increased welding efficiency, suggest that this new technology should be considered for radioactive materials packaging campaigns. FSW technology will require some development, adaptation for this application, along with several activities needed for commercialization. One of these activities will be to obtain approval from the governing construction code to use the FSW technology. The American Society of Mechanical Engineers Boiler and Pressure Vessel Code (ASME B&PVC) will govern this work; however, rules for the use of FSW are not currently addressed. A code case will be required, defining appropriate process variables within prescribed limits, and submitted to the Code for review/approval and incorporation.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Three-dimensional Zn-based alloys for dendrite-free aqueous Zn battery in dual-cation electrolytes

Aqueous zinc-ion batteries, in terms of integration with high safety, environmental benignity, and low cost, have attracted much attention for powering electronic devices and storage systems. However, the interface instability issues at the Zn anode caused by detrimental side reactions such as dendrite growth, hydrogen evolution, and metal corrosion at the solid (anode)/liquid (electrolyte) interface impede their practical applications in the fields requiring long-term performance persistence. Despite the rapid progress in suppressing the side reactions at the materials interface, the mechanism of ion storage and dendrite formation in practical aqueous zinc-ion batteries with dual-cation aqueous electrolytes is still unclear. Herein, we design an interface material consisting of forest-like three-dimensional zinc-copper alloy with engineered surfaces to explore the Zn plating/stripping mode in dual-cation electrolytes. The three-dimensional nanostructured surface of zinc-copper alloy is demonstrated to be in favor of effectively regulating the reaction kinetics of Zn plating/stripping processes. The developed interface materials suppress the dendrite growth on the anode surface towards high-performance persistent aqueous zinc-ion batteries in the aqueous electrolytes containing single and dual cations. This work remarkably enhances the fundamental understanding of dual-cation intercalation chemistry in aqueous electrochemical systems and provides a guide for exploring high-performance aqueous zinc-ion batteries and beyond.

25 ENERGY STORAGE↗

Development of Tailored Hydrocarbon-Based Pentablock Copolymer Membranes for Sodium-Polysulfide Flow Batteries

Long-duration energy storage (LDES) technologies are pivotal for the adoption of renewables like wind and solar. Non-aqueous redox flow batteries (NARFBs) with a sodium-polysulfide hybrid system feature high energy density independent of power density, yet face challenges with polysulfide shuttling. This study investigates a hydrocarbon-based penta-block copolymer membrane, Nexar, to mitigate crossover effects by balancing TFSI conversion and their crosslink density. The membranes are annealed to induce crosslinking for reducing electrolyte uptake and enhancing mechanical stability while demonstrating excellent ionic conductivity. The hydrocarbon-based membranes address environmental concerns associated with perfluoroalkyl substances and improve the performance and durability of NARFBs. In conclusion, our findings suggest that annealed Nexar membranes with tailored TFSI functionality offer a scalable, cost-effective solution for enhancing the efficiency of high-capacity energy storage systems, pivotal for grid integration of renewable sources.

25 ENERGY STORAGE↗

Ferroelectric opening switches for large-scale pulsed power drivers.

Fast electrical energy storage or Voltage-Driven Technology (VDT) has dominated fast, high-voltage pulsed power systems for the past six decades. Fast magnetic energy storage or Current-Driven Technology (CDT) is characterized by 10,000 X higher energy density than VDT and has a great number of other substantial advantages, but it has all but been neglected for all of these decades. The uniform explanation for neglect of CDT technology is invariably that the industry has never been able to make an effective opening switch, which is essential for the use of CDT. Most approaches to opening switches have involved plasma of one sort or another. On a large scale, gaseous plasmas have been used as a conductor to bridge the switch electrodes that provides an opening function when the current wave front propagates through to the output end of the plasma and fully magnetizes the plasma - this is called a Plasma Opening Switch (POS). Opening can be triggered in a POS using a magnetic field to push the plasma out of the A-K gap - this is called a Magnetically Controlled Plasma Opening Switch (MCPOS). On a small scale, depletion of electron plasmas in semiconductor devices is used to affect opening switch behavior, but these devices are relatively low voltage and low current compared to the hundreds of kilo-volts and tens of kilo-amperes of interest to pulsed power. This work is an investigation into an entirely new approach to opening switch technology that utilizes new materials in new ways. The new materials are Ferroelectrics and using them as an opening switch is a stark contrast to their traditional applications in optics and transducer applications. Emphasis is on use of high performance ferroelectrics with the objective of developing an opening switch that would be suitable for large scale pulsed power applications. Over the course of exploring this new ground, we have discovered new behaviors and properties of these materials that were here to fore unknown. Some of these unexpected discoveries have lead to new research directions to address challenges.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Technoeconomic Analysis of Discrete and Unitized Reversible Fuel Cells for Energy Storage Applications

Reversible Fuel Cell (RFC) systems offer promising characteristics for stationary long duration energy storage applications. Two main configurations of RFC systems exist: discrete RFC systems and unitized RFC systems. While discrete RFC systems combine independent fuel cell and electrolyzer systems for energy storage, unitized RFC systems utilize a single electrochemical stack and might share balance of plant (BOP) components for both charging and discharging processes. While this configuration reduces upfront capital costs, challenges of unitized RFC designs include potential performance trade-offs due to dual-mode stack design and operational complexities across varying loads and operating conditions. Furthermore, these tradeoffs might be different for low-temperature PEM RFCs than for high-temperature reversible solid oxide cell systems. The goal of this project is to assess unitized RFC system potential in the context of long duration grid energy storage and HFTO's technical targets for different discrete fuel cell and electrolyzer technologies. This presentation presents preliminary review of state-of-the-art unitized RFC cells and assesses how they perform relative to HFTO's technical targets. It also presents literature-derived system configurations worth investigating. This review indicates that lab-scale unitized RFCs are making good progress towards meeting HFTO's technical targets.

HYDROGEN↗

A comparative technoeconomic analysis of renewable hydrogen production using solar energy

A technoeconomic analysis of photoelectrochemical (PEC) and photovoltaic-electrolytic (PV-E) solar-hydrogen production of 10 000 kg H 2 day -1 (3.65 kilotons per year) was performed to assess the economics of each technology, and to provide a basis for comparison between these technologies as well as within the broader energy landscape. Two PEC systems, differentiated primarily by the extent of solar concentration (unconcentrated and 10× concentrated) and two PV-E systems, differentiated by the degree of grid connectivity (unconnected and grid supplemented), were analyzed. In each case, a base-case system that used established designs and materials was compared to prospective systems that might be envisioned and developed in the future with the goal of achieving substantially lower overall system costs. With identical overall plant efficiencies of 9.8%, the unconcentrated PEC and non-grid connected PV-E system base-case capital expenses for the rated capacity of 3.65 kilotons H 2 per year were 205 dollars MM (293 dollars per m 2 of solar collection area (m S -2 ), 14.7 W H2,P -1 ) and 260 dollars MM ($371 m S -2 , 18.8 dollars W H2,P -1 ), respectively. The untaxed, plant-gate levelized costs for the hydrogen product (LCH) were $11.4 kg -1 and 12.1 dollars kg -1 for the base-case PEC and PV-E systems, respectively. The 10× concentrated PEC base-case system capital cost was 160 dollars MM (428 dollars m S -2 , 11.5 dollars W H2,P -1 ) and for an efficiency of 20% the LCH was 9.2 kg -1 . Likewise, the grid supplemented base-case PV-E system capital cost was 66 dollars MM (441 dollars m S -2 , 11.5 dollars W H2,P -1 ), and with solar-to-hydrogen and grid electrolysis system efficiencies of 9.8% and 61%, respectively, the LCH was 6.1 dollars kg -1 . As a benchmark, a proton-exchange membrane (PEM) based grid-connected electrolysis system was analyzed. Assuming a system efficiency of 61% and a grid electricity cost of $0.07 kWh -1 , the LCH was $5.5 kg -1 . A sensitivity analysis indicated that, relative to the base-case, increases in the system efficiency could effect the greatest cost reductions for all systems, due to the areal dependencies of many of the components. The balance-of-systems (BoS) costs were the largest factor in differentiating the PEC and PV-E systems. No single or combination of technical advancements based on currently demonstrated technology can provide sufficient cost reductions to allow solar hydrogen to directly compete on a levelized cost basis with hydrogen produced from fossil energy. Specifically, a cost of CO 2 greater than ~$800 dollars (ton CO 2 ) -1 was estimated to be necessary for base-case PEC hydrogen to reach price parity with hydrogen derived from steam reforming of methane priced at $12 GJ -1 ($1.39 (kg H 2 ) -1 ). A comparison with low CO 2 and CO 2 -neutral energy sources indicated that base-case PEC hydrogen is not currently cost-competitive with electrolysis using electricity supplied by nuclear power or from fossil-fuels in conjunction with carbon capture and storage. Solar electricity production and storage using either batteries or PEC hydrogen technologies are currently an order of magnitude greater in cost than electricity prices with no clear advantage to either battery or hydrogen storage as of yet. Significant advances in PEC technology performance and system cost reductions are necessary to enable cost-effective PEC-derived solar hydrogen for use in scalable grid-storage applications as well as for use as a chemical feedstock precursor to CO 2 -neutral high energy-density transportation fuels. Hence such applications are an opportunity for foundational research to contribute to the development of disruptive approaches to solar fuels generation systems that can offer higher performance at much lower cost than is provided by current embodiments of solar fuels generators. Efforts to directly reduce CO 2 photoelectrochemically or electrochemically could potentially produce products with higher value than hydrogen, but many, as yet unmet, challenges include catalytic efficiency and selectivity, and CO 2 mass transport rates and feedstock cost. Major breakthroughs are required to obtain viable economic costs for solar hydrogen production, but the barriers to achieve cost-competitiveness with existing large-scale thermochemical processes for CO 2 reduction are even greater.

14 SOLAR ENERGY↗

Multi-Functional Distributed Fiber Sensors for Pipeline Monitoring and Methane Detections. Final Report

As an abundant and cheap fossil energy source, natural gas has become a significant energy supply to support the United States’ economy. However, the large-scale extraction and utilization of natural gas also impose significant challenges on methane leakage. This problem is exacerbated by aging gas utility delivery systems, including interstate high-pressure pipelines, storage, and transmission facilities. This project aims to develop a cost-effective fiber optical sensing method that can perform multi-parameter real-time measurements of natural gas pipelines across long interrogation distances up to 100 km with 1-meter spatial resolution. This sensing tool can evaluate overall pipeline efficiency and reduce methane emissions for mid-stream methane infrastructures. To accomplish this objective, research and development efforts funded by this project have resulted in the following accomplishments: This project successfully has developed new functional sensory polymer materials using Metal-Organic Frameworks (MOFs) that can be coated on optical fiber through the reel-to-reel coating process. Functional polymer-coated optical fibers can perform sensitive methane detection through evanescence wave interaction and strain-based measurements to achieve 1% detection sensitivities. The new sensors fibers support both distributed measurements and multiplexed fiber sensors array for multi-point measurements. This project developed and optimized a new multi-core optical fiber that supports simultaneous and distributed measurements of strain and temperatures with 1-meter spatial resolutions across up to 100-km interrogation distance. This new fiber, combined with sensory polymercoated fiber, could perform both distributed temperature and methane detections. This project developed a new artificial intelligence big data algorithm approach that can effectively analyze high-resolution data harnessed by distributed fiber sensors to protect natural gas pipelines against external threats and detect internal defects induced by corrosion. Working with our industry partner, this project developed new optical fibers that support fiber sensor fabrications through polymer coating after the fibers are drawn. These new fibers eliminate the need for direct sensor fabrication when the fiber is fabricated on a fiber draw tower, which drastically expands fiber sensors' applicability. This research project has significantly advanced the distributed fiber sensing technology. It will dramatically increase the applicability and adaptability of distributed fiber sensors for a wide array of applications in energy, sustainability, and environmental science, including structural health monitoring of natural gas pipelines, oil infrastructures, hydrogen facilities, and environmental monitoring of carbon storage sites, water supply systems, and others.

03 NATURAL GAS↗

National Center of Excellence for Energy Storage Technology 168.10 (Final Report for Period 10/1/2010-12/31/2011)

This report documents the performance of the Ohio State University (OSU) and Edison Welding Institute (EWI) in the period from 10/1/2010 to 12/31/2012. The objective of the project is to establish a Center of Excellence that leverages the strengths of the partners to establish a unique capability to develop and transfer energy storage industries to establish a unique capability in the development and transfer of energy storage system technology through a fundamental understanding of battery electrical and thermal performance, damage and aging mechanisms, and through the development of reliable, high-speed processes for joining substrates in battery cell, module and pack assemblies with low manufacturing variability. During this period, the OSU activity focused on procuring the equipment, materials and supplies necessary to conduct the experiments planned in the statement of project objectives. In detail, multiple laboratory setups were developed to enable for characterizing the open-circuit potential of cathode and anode materials for Li-ion batteries, perform experiments on calorimetry, and finally built multiple cell and module battery cyclers to be able to perform aging campaign on a wide variety of automotive grade battery cells and small modules. This suite of equipment feeds directly into the development, calibration of battery models ranging from first principle electrochemical models to electro-thermal equivalent circuit models suitable for use in control and xEV vehicle simulations. In addition, it allows to develop and calibrate ‘aging’ models for Li-ion batteries that enable the development of diagnostics and prognostics tools to characterize and predict battery degradation from automotive usage under a wide array of environmental and usage scenarios. The objective of the EWI work scope is to develop improved processes for making metal-tometal joints in advanced battery cells and packs. It will focus on developing generic techniques for making functional (electrically conductive and mechanically robust) metal-to-metal joints between thin substrates. Joints with multiple layers and bimetallic constituents will be investigated. During the current period of performance, EWI has defined the test matrix to evaluate the application of different welding technologies (laser welding, ultrasonic welding, resistance welding) to specific components of battery cells and modules, such as foils-to-tabs, tabs-to-tabs, and tabs-to-bus bars. The test matrix also includes a range of substrates (aluminum 1145 and 1100, copper 110 and nickel 200 as substrates). Furthermore, a set of procedures was defined to perform mechanical and electrical testing of the samples, including metallography, and non-destructive evaluations. Both on the OSU and EWI, this project enabled to leverage very significant industrial collaborations with a wide array of companies ranging from battery manufacturers and pack integrator all the ways to Tier 1 automotive suppliers and OEMs during the period of exercise of the project, and in the future for years to come.

25 ENERGY STORAGE↗

Design and Modeling of a Prototype Floating Membrane Reservoir System Application for Pumped Storage Hydropower

Design and modeling were performed for a prototype floating membrane reservoir system for closed-loop pumped storage hydropower (PSH) application. Typically, new construction of PSH requires high up-front development costs and long investment return periods. The proposed floating membrane reservoir system innovative technology would eliminate the need for highly disruptive and costly reservoir construction. Such modular systems may ultimately provide reliable, flexible power generation and can be replicated multiple times to increase energy storage capacity. This paper documents the modeling response analysis completed to refine and validate the design prior to the testing of the prototype. Computational fluid dynamics modeling efforts were completed to assess hydrostatic and hydrodynamic forces acting on the reservoir for several deployment and alignment arrangements. The system stability was evaluated using material stress and frequency response analysis. The conclusions, based on hydrodynamic and structural response modeling, support a decision to move forward with construction, deployment, and testing of the innovative prototype floating reservoir system.

Hadjerioua, Boualem↗

Heterogeneous current collector in lithium-ion battery for thermal-runaway mitigation

Current collector accounts for more than 90% of the electric conductivity and ∼90% of the mechanical strength of the electrode in lithium-ion battery (LIB). Usually, current collectors are smooth metallic thin films. In the current study, we show that if the current collector is heterogeneous, the heat generation becomes negligible when the LIB cell is subjected to mechanical abuse. The phenomenon is attributed to the guided strain concentration, which promotes the separation of the forward and the return paths of internal short circuit. As the internal impedance drastically increases, the stored electric energy cannot be dissipated as thermal energy. The modification of current collector does not affect the cycling performance of the LIB cell. This finding enables advanced thermal-runaway mitigation techniques for high-energy, large-scale energy storage systems.

Wang, Meng↗

I/O in Machine Learning Applications on HPC Systems: A 360-degree Survey

Growing interest in Artificial Intelligence (AI) has resulted in a surge in demand for faster methods of Machine Learning (ML) model training and inference. This demand for speed has prompted the use of high performance computing (HPC) systems that excel in managing distributed workloads. Because data is the main fuel for AI applications, the performance of the storage and I/O subsystem of HPC systems is critical. In the past, HPC applications accessed large portions of data written by simulations or experiments or ingested data for visualizations or analysis tasks. ML workloads perform small reads spread across a large number of random files. This shift of I/O access patterns poses several challenges to modern parallel storage systems. In this paper, we survey I/O in ML applications on HPC systems, and target literature within a 6-year time window from 2019 to 2024. We define the scope of the survey, provide an overview of the common phases of ML, review available profilers and benchmarks, examine the I/O patterns encountered during offline data preparation, training, and inference, and explore I/O optimizations utilized in modern ML frameworks and proposed in recent literature. Lastly, we seek to expose research gaps that could spawn further R&D.

97 MATHEMATICS AND COMPUTING↗

R&D100: Lightweight Distributed Metric Service

On today's High Performance Computing platforms, the complexity of applications and configurations makes efficient use of resources difficult. The Lightweight Distributed Metric Service (LDMS) is monitoring software developed by Sandia National Laboratories to provide detailed metrics of system performance. LDMS provides collection, transport, and storage of data from extreme-scale systems at fidelities and timescales to provide understanding of application and system performance with no statistically significant impact on application performance.

97 MATHEMATICS AND COMPUTING↗