SEARCH · Engineering Papers

Results for “latency”

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.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 37 records · Page 2

All sources NASA NTRS DOE OSTI

NASA NTRS2011

Low-Latency Lunar Surface Telerobotics from Earth-Moon Libration Points

Concepts for a long-duration habitat at Earth-Moon LI or L2 have been advanced for a number of purposes. We propose here that such a facility could also have an important role for low-latency telerobotic control of lunar surface equipment, both for lunar science and development. With distances of about 60,000 km from the lunar surface, such sites offer light-time limited two-way control latencies of order 400 ms, making telerobotic control for those sites close to real time as perceived by a human operator. We point out that even for transcontinental teleoperated surgical procedures, which require operational precision and highly dexterous manipulation, control latencies of this order are considered adequate. Terrestrial telerobots that are used routinely for mining and manufacturing also involve control latencies of order several hundred milliseconds. For this reason, an Earth-Moon LI or L2 control node could build on the technology and experience base of commercially proven terrestrial ventures. A lunar libration-point telerobotic node could demonstrate exploration strategies that would eventually be used on Mars, and many other less hospitable destinations in the solar system. Libration-point telepresence for the Moon contrasts with lunar telerobotic control from the Earth, for which two-way control latencies are at least six times longer. For control latencies that long, telerobotic control efforts are of the "move-and-wait" variety, which is cognitively inferior to near real-time control.

Lester, Daniel↗

NASA NTRS2016

Analog Testing of Operations Concepts for Mitigation of Communication Latency During Human Space Exploration

OBJECTIVES: NASA Extreme Environment Mission Operations (NEEMO) is an underwater spaceflight analog that allows a true mission‐like operational environment and uses buoyancy effects and added weight to simulate different gravity levels. Three missions were undertaken from 2014‐2015, NEEMO's 18‐20. All missions were performed at the Aquarius undersea research habitat. During each mission, the effects of varying operations concepts and tasks type and complexity on representative communication latencies associated with Mars missions were studied. METHODS: 12 subjects (4 per mission) were weighed out to simulate near‐zero or partial gravity extravehicular activity (EVA) and evaluated different operations concepts for integration and management of a simulated Earth‐based science backroom team (SBT) to provide input and direction during exploration activities. Exploration traverses were planned in advance based on precursor data collected. Subjects completed science‐related tasks including presampling surveys, geologic‐based sampling, and marine‐based sampling as a portion of their tasks on saturation dives up to 4 hours in duration that were to simulate extravehicular activity (EVA) on Mars or the moons of Mars. One‐way communication latencies, 5 and 10 minutes between space and mission control, were simulated throughout the missions. Objective data included task completion times, total EVA times, crew idle time, translation time, SBT assimilation time (defined as time available for SBT to discuss data/imagery after it has been collected, in addition to the time taken to watch imagery streaming over latency). Subjective data included acceptability, simulation quality, capability assessment ratings, and comments. RESULTS: Precursor data can be used effectively to plan and execute exploration traverse EVAs (plans included detailed location of science sites, high‐fidelity imagery of the sites, and directions to landmarks of interest within a site). Operations concepts that allow for presampling surveys enable efficient traverse execution and meaningful Mission Control Center (MCC) interaction across long communication latencies and can be done with minimal crew idle time. Imagery and information from the EVA crew that is transmitted real‐time to the intravehicular (IV) crewmember(s) can be used to verify that exploration traverse plans are being executed correctly. That same data can be effectively used by MCC (across comm latency) to provide further instructions to the crew from a SBT on sampling priorities, additional tasks, and changes to the plan. Text / data capabilities are preferred over voice capabilities between MCC and IV when executing exploration traverse plans over communication latency. Autonomous crew planning tools can be effective at modifying existing plans if the objectives and constraints are clearly defined.

Chappell, Steven P.↗

DOE OSTI2026

A Flexible Forwarding Scheme to Improve Latency-Bound Irregular P2P Communication in MPI

We propose an algorithm to efficiently perform latency-bound communication scenarios that consist of many small messages. In these parallel scenarios, processes typically pass around a lot of small-sized messages of a few KBs of size. Performing communication operations with P2P MPI routines or collective MPI routines (including neighborhood collectives) in such scenarios may not always yield the optimal results and may not resolve the latency bottleneck. To this end, we develop a regular structure called virtual process topology (VPT) on which the messages can be communicated in a structured and controlled manner. Using parameters of this topology, one can tune the rate of aggression in tackling the latency costs. We demonstrate that our communication algorithm is preferable to MPI P2P and collective routines for latency-bound communication and it can easily be adapted only by replacing calls to MPI routines in a parallel application. We show how to adapt existing topology-aware mapping heuristics to address the volume overhead due to communicating messages on the VPT. Moreover, we propose a novel swap-based mapping heuristic to address this overhead by optimizing the maximum volume handled by a process. Experiments on synthetic communication graphs as well as real-world applications such as parallel Canonical Polyadic sparse tensor decomposition and parallel sparse matrix-dense matrix multiplication show that our approach is a powerful way of overcoming the bottlenecks posed by sparse and latency-bound irregular communication.

communication algorithm↗

DOE OSTI2026

Latency Analysis of the Nexus Digital Twin Framework

Real-time digital catalogs are increasingly relied upon to track metadata and connect disparate data sources for cloud-based data integration efforts. One such tool, Deeplynx Nexus is supporting real-time digital twin efforts through event-driven data integration and time-series queries. Nexus’s usefulness for these applications depends critically on how quickly individual records can be uploaded and downloaded, since delays directly affect the responsiveness of any system built on top of it. However, the actual latency a user should expect from Nexus has not been systematically measured before, particularly for the small, frequent transactions typical of live sensor feeds. Here we show that single-record round-trip latency is 61.1 ms on a local Nexus instance and 391.7 ms on the hosted production infrastructure, a roughly 6.4x difference driven primarily by fixed per-request overhead rather than data volume. This overhead dominates at small scale: comparing single-record and ten-record trials suggests approximately 56 ms of each single-record request is fixed connection and authentication cost rather than data-transfer time, meaning batching even a handful of records is substantially more efficient than transmitting them individually. At large batch sizes, this pattern reverses for uploads, which converge to near parity between local and hosted environments by 25,000-50,000 records, while download latency remains persistently 5.7-6.4x slower on hosted infrastructure even at scale. These results suggest that Nexus deployments intended for real-time digital twin applications should prioritize record batching over single-record transactions, and that download-path optimization on hosted infrastructure offers the largest remaining opportunity to reduce latency at scale. We anticipate these baseline measurements will serve as a reference point for future digital twin projects evaluating whether Nexus’s latency profile meets their real-time requirements, and as a benchmark for tracking the effect of future infrastructure or API changes.

99 - GENERAL AND MISCELLANEOUS↗

DOE OSTI2023

Recombination smooths the time-signal disrupted by latency in within-host HIV phylogenies

Within-host HIV evolution involves several features that may disrupt standard phylogenetic reconstruction. One important feature is re-activation of latently integrated provirus, which has the potential to disrupt the temporal signal, leading to variation in the branch lengths and apparent evolutionary rates in a tree. Yet, real within-host HIV phylogenies tend to show clear, ladder-like trees structured by the time of sampling. Another important feature is recombination, which violates the fundamental assumption that evolutionary history can be represented by a single bifurcating tree. Thus, recombination complicates the within-host HIV dynamic by mixing genomes and creating evolutionary loop structures that cannot be represented in bifurcating trees. In this paper, we develop a coalescent-based simulator of within-host HIV evolution that includes latency, recombination, and effective population size dynamics that allows us to study the relationship between the true, complex genealogy of within-host HIV evolution, encoded as an Ancestral Recombination Graph (ARG), and the observed phylogenetic tree. To compare our ARG results to the familiar phylogeny format, we calculate the expected bifurcating tree after decomposing the ARG into all unique site trees, their combined distance matrix, and the overall corresponding bifurcating tree. While latency and recombination separately disrupt the phylogenetic signal, remarkably, we find that recombination recovers the temporal signal of within-host HIV evolution caused by latency by mixing fragments of old, latent genomes into the contemporary population. In effect, recombination averages over extant heterogeneity, whether it stems from mixed time-signals or population bottlenecks. Further, we establish that the signals of latency and recombination can be observed in phylogenetic trees despite being an incorrect representation of the true evolutionary history. Using an Approximate Bayesian Computation method, we develop a set of statistical probes to tune our simulation model to nine longitudinally-sampled within-host HIV phylogenies. Because ARGs are exceedingly difficult to infer from real HIV data, our simulation system allows investigating effects of latency, recombination, and population size bottlenecks by matching decomposed ARGs to real data as observed in standard phylogenies.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2025

wa-hls4ml and lui-gnn: A benchmark and GNN-based surrogate model for hls4ml resource and latency estimation

As machine learning (ML) increasingly serves as a tool for addressing real-time challenges in scientific applications, the development of advanced tooling has significantly reduced the time required to iterate on various designs. These advancements have solved major obstacles, but also exposed new challenges. For example, processes that were not previously considered bottlenecks, such as model synthesis, are now becoming limiting factors in the rapid iteration of designs. To reduce these emerging constraints, multiple efforts are being launched toward designing an ML-based surrogate model that estimates resource usage of synthesized accelerator architectures. This model would reduce the design iteration time, especially when designing within a set of given hardware constraints. This approach shows considerable potential, but as it stands, the effort is early and would benefit from coordination and standardization to assist future work as it emerges. We introduce wa-hls4ml, a benchmark for ML accelerator resource and latency estimation, and its corresponding initial dataset of more than 100,000 fully connected neural networks, all synthesized using hls4ml and targeting Xilinx FPGAs. In addition to the resource utilization and latency data provided, the dataset includes generated artifacts and log files for many of the synthesized neural networks, in order to support future research in ML-based code generation. The benchmark evaluates the performance of resource and latency predictors against several common ML model architectures, primarily originating from scientific domains, as exemplar models, as well as the average performance across a subset of the dataset. We measure the performance of a given predictor model through multiple metrics, including $R^2$ score and SMAPE on regression tasks, as well as inference time to further characterize the estimator under test. Additionally, we introduce the latency/utilization inference graph neural network (lui-gnn), a surrogate model that uses a graph neural network to represent input architectures in the form of a directed graph. This graph representation allows for a diverse set of model architectures to all be effectively handled by a surrogate model. We present the architecture and performance of the model, as evaluated by the new proposed benchmark, including SMAPE, $R^2$ score, and inference times, and find that lui-gnn generally predicts latency and utilization for the 75\% quantile within several percent of the synthesized resources on the synthetic test dataset, indicating that this approach of estimating resource and latency via a surrogate models has promise and warrants further research.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

NASA NTRS1989

An experimental study of memory fault latency

The difficulty with the measurement of fault latency is due to the lack of observability of the fault occurrence and error generation instants in a production environment. The authors describe an experiment, using data from a VAX 11/780 under real workload, to study fault latency in the memory subsystem accurately. Fault latency distributions are generated for stuck-at-zero (s-a-0) and stuck-at-one (s-a-1) permanent fault models. The results show that the mean fault latency of an s-a-0 fault is nearly five times that of the s-a-1 fault. An analysis of variance is performed to quantify the relative influence of different workload measures on the evaluated latency.

Chillarege, Ram↗

NASA NTRS1997

Eye Movement Latencies to Direction Change for Different Classes of Motion

In the analysis of visual motion, local features such as orientation are analyzed early in the cortical processing stream (V1), while integration across orientation and space is thought to occur in higher cortical areas such as MT, MST, etc. If all areas provide inputs to eye movement control centers, we would expect that local properties would drive eye movements with relatively short latencies, while global properties would require longer latencies. When such latencies are observed, they can provide information about when (and where?) various stimulus properties are analyzed. To this end, a stimulus was employed in which local and global properties determining perceived direction-of-motion could be manipulated independently: an elliptical Gabor patch with a drifting carrier, with variable orientation of the carrier grating and the contrast window. We have previously demonstrated that the directional percepts evoked by this stimulus vary between the "grating direction" (the normal to the grating's orientation) and the "window direction" (ARVO 91, 94), and that similar effects can be observed in reflexive eye movements (ARVO 95). Subjects viewed such a stimulus while attempting to maintain steady fixation on the center of the pattern, and the small reflexive eye movements ("stare OKN") were recorded. In the middle of the trial, the orientation of either the grating or the window was rotated smoothly by 30 degrees. Responses to the shift of both grating orientation and window orientation are seen in the average OKN slow phase velocity. Grating rotations produce a rapid OKN rotation to the grating direction (100 ms latency, 300 ms time constant), followed by a slower rebound to the steady state perceived direction midway between the grating and window directions. Window rotations, on the other hand, evoke a slower response (200 ms latency, 500 ms time constant). The results demonstrate multiple cortical inputs to eye movement control: a taste early input driven by orientation, and a slower input from higher areas sensitive to global stimulus properties.

Mulligan, Jeffrey B.↗

NASA NTRS2005

Latency in Visionic Systems: Test Methods and Requirements

A visionics device creates a pictorial representation of the external scene for the pilot. The ultimate objective of these systems may be to electronically generate a form of Visual Meteorological Conditions (VMC) to eliminate weather or time-of-day as an operational constraint and provide enhancement over actual visual conditions where eye-limiting resolution may be a limiting factor. Empirical evidence has shown that the total system delays or latencies including the imaging sensors and display systems, can critically degrade their utility, usability, and acceptability. Definitions and measurement techniques are offered herein as common test and evaluation methods for latency testing in visionics device applications. Based upon available data, very different latency requirements are indicated based upon the piloting task, the role in which the visionics device is used in this task, and the characteristics of the visionics cockpit display device including its resolution, field-of-regard, and field-of-view. The least stringent latency requirements will involve Head-Up Display (HUD) applications, where the visionics imagery provides situational information as a supplement to symbology guidance and command information. Conversely, the visionics system latency requirement for a large field-of-view Head-Worn Display application, providing a Virtual-VMC capability from which the pilot will derive visual guidance, will be the most stringent, having a value as low as 20 msec.

Bailey, Randall E.↗

NASA NTRS2016

NEEMO 18-20: Analog Testing for Mitigation of Communication Latency During Human Space Exploration

NASA Extreme Environment Mission Operations (NEEMO) is an underwater spaceflight analog that allows a true mission-like operational environment and uses buoyancy effects and added weight to simulate different gravity levels. Three missions were undertaken from 2014-2015, NEEMO's 18-20. All missions were performed at the Aquarius undersea research habitat. During each mission, the effects of communication latencies on operations concepts, timelines, and tasks were studied. METHODS: Twelve subjects (4 per mission) were weighed out to simulate near-zero or partial gravity extravehicular activity (EVA) and evaluated different operations concepts for integration and management of a simulated Earth-based science team (ST) to provide input and direction during exploration activities. Exploration traverses were preplanned based on precursor data. Subjects completed science-related tasks including pre-sampling surveys, geologic-based sampling, and marine-based sampling as a portion of their tasks on saturation dives up to 4 hours in duration that were designed to simulate extravehicular activity (EVA) on Mars or the moons of Mars. One-way communication latencies, 5 and 10 minutes between space and mission control, were simulated throughout the missions. Objective data included task completion times, total EVA times, crew idle time, translation time, ST assimilation time (defined as time available for ST to discuss data/imagery after data acquisition). Subjective data included acceptability, simulation quality, capability assessment ratings, and comments. RESULTS: Precursor data can be used effectively to plan and execute exploration traverse EVAs (plans included detailed location of science sites, high-fidelity imagery of the sites, and directions to landmarks of interest within a site). Operations concepts that allow for pre-sampling surveys enable efficient traverse execution and meaningful Mission Control Center (MCC) interaction across communication latencies and can be done with minimal crew idle time. Imagery and contextual information from the EVA crew that is transmitted real-time to the intravehicular (IV) crewmember(s) can be used to verify that exploration traverse plans are being executed correctly. That same data can be effectively used by MCC (across comm latency) to provide meaningful feedback and instruction to the crew regarding sampling priorities, additional tasks, and changes to the EVA timeline. Text / data capabilities are preferred over voice capabilities between MCC and IV when executing exploration traverse plans over communication latency.

Chappell, Steven P.↗

NASA NTRS2017

A Simulation Base Investigation of High Latency Space Systems Operations

NASA's human space program has developed considerable experience with near Earth space operations. Although NASA has experience with deep space robotic missions, NASA has little substantive experience with human deep space operations. Even in the Apollo program, the missions lasted only a few weeks and the communication latencies were on the order of seconds. Human missions beyond the relatively close confines of the Earth-Moon system will involve missions with durations measured in months and communications latencies measured in minutes. To minimize crew risk and to maximize mission success, NASA needs to develop a better understanding of the implications of these types of mission durations and communication latencies on vehicle design, mission design and flight controller interaction with the crew. To begin to address these needs, NASA performed a study using a physics-based subsystem simulation to investigate the interactions between spacecraft crew and a ground-based mission control center for vehicle subsystem operations across long communication delays. The simulation, built with a subsystem modeling tool developed at NASA's Johnson Space Center, models the life support system of a Mars transit vehicle. The simulation contains models of the cabin atmosphere and pressure control system, electrical power system, drinking and waste water systems, internal and external thermal control systems, and crew metabolic functions. The simulation has three interfaces: 1) a real-time crew interface that can be use to monitor and control the vehicle subsystems; 2) a mission control center interface with data transport delays up to 15 minutes each way; 3) a real-time simulation test conductor interface that can be use to insert subsystem malfunctions and observe the interactions between the crew, ground, and simulated vehicle. The study was conducted at the 21st NASA Extreme Environment Mission Operations (NEEMO) mission between July 18th and Aug 3rd of year 2016. The NEEMO mission provides ideal conditions for this study with crew in the loop, an active control center, and real-time flow of high latency communications and data. NEEMO crew and ground support will work through procedures including activation of the transit vehicle power system, opening the hatch between the transit vehicle and a Mars ascent vehicle, transferring simulated crewmembers between vehicles, overcoming subsystem malfunctions, sending simulated crewmember on extra-vehicular activities, and other housekeeping activities. This study is enhancing the understanding of high latency operations and the advantages and disadvantages of different communication methods. It is also providing results that will help improve the design of simulation interfaces and inform the design of Mars transit vehicles.

Li, Zu Qun↗

NASA NTRSDate not supplied

Why Near Real-time/Low Latency is Important to Monitor the Changing World

An essential factor for remote sensing data products in impacting decision making is latency, or the time between earth observation and data are available to users. In many applications areas, latency plays an important or even decisive role where low latency earth observations help people to make timely, data-based decisions. Within the open and free NASA resources, NASA’s Land, Atmosphere Near real-time Capability for Earth Observing System (EOS) (LANCE) supports users interested in monitoring a wide variety of natural and man-made phenomena using near real-time (NRT) data products that are made available much quicker than routine processing allows. The combination of all available LANCE satellite products provides global coverage at multiple times per day, which makes it possible to meet user needs in various areas of applications including water resources, agriculture, air quality, wildland fire and many other disasters monitoring and management. As one of the prime users of LANCE, NASA’s Earth Science Applied Sciences Program promotes the use of LANCE data products to demonstrate applications in decision making and facilitates the end-user feedback to the science team to improve data products. One of the most critical applications that we expect data to be processed as close to the user as possible is wildland file response and management. User feedback indicates that LANCE NRT fire data products within 3 hours latency would meet the needs of the wildland fire community. Other examples of earth science application areas for which low latency is particularly important include detecting volcanic eruptions, early warning of disasters, tracking extreme weather events, and monitoring air quality.

Tian Yao↗

NASA NTRSDate not supplied

The Impact and Usage of Near Real-Time and Low Latency Solutions from the Satellite Needs Working Group

The Satellite Needs Working Group (SNWG) is an interagency effort of the U.S. Government to identify the Earth observation gaps and data needs across civilian U.S. federal agencies. Within short time frames of 2 to 5 years, on behalf of SNWG, NASA formulates and implements a range of remote sensing-based solutions that benefit the entire Earth Science community and society at large. More than 20 such solutions are underway, with several addressing needs for near real-time and low latency information as a result of priorities identified in the survey. In this talk, we will present and discuss the impact and usage of such solutions, including ICESat-2 Atlas products (data delivered within 72 hours instead of the standard 45-day latency for ATLAS science products), near real-time air quality TEMPO products derived from the recently launched TEMPO mission (these products include SO 2 , NO 2 , HCHO, O 2 -O 2 cloud cover, and aerosols with a planned latency of one hour), and low latency Harmonized Landsat-Sentinel-2 (HLS) data (HLS data with a latency of 6 hours). We will talk about how these solutions are being co-designed by federal agencies to maximize their impact and usage by the scientific community. Finally, we will also invite additional stakeholders and users to get involved in the formulation, implementation, and operationalization of these solutions.

Pontus Olofsson↗

NASA NTRS2014

User Needs and Assessing the Impact of Low Latency NASA Earth Observation Data Availability on Societal Benefit

Since the advent of NASA's Earth Observing System, knowledge of the practical benefits of Earth science data has grown considerably. The community using NASA Earth science observations in applications has grown significantly, with increasing sophistication to serve national interests. Data latency, or how quickly communities receive science observations after acquisition, can have a direct impact on the applications and usability of the information. This study was conducted to determine how users are incorporating NASA data into applications and operational processes to benefit society beyond scientific research, as well as to determine the need for data latency of less than 12 h. The results of the analysis clearly show the significant benefit to society of serving the needs of the agricultural, emergency response, environmental monitoring and weather communities who use rapidly delivered, accurate Earth science data. The study also showed the potential of expanding the communities who use low latency NASA science data products to provide new ways of transforming data into information. These benefits can be achieved with a clear and consistent NASA policy on product latency.

NASA↗

DOE OSTI2021

Scalable Distributed Reachability Analysis for Cyber-Physical Networked Microgrids with Communication Latency

Due to the high penetration of power-electronic-interfaced distributed energy resources (DERs), a microgrid's inertia is significantly reduced, making it sensitive to cyberand physical operational changes. When multiple microgrids are interconnected to form networked microgrids for improving system's resilience, the changes of one microgrid can quickly escalate to the overall system. In cyber-physical networked microgrids, the communication latency varies a lot. To efficiently quantify the impact of the communication latency on the dynamics of networked microgrids, a scalable distributed reachability analysis approach is presented in this paper. Reachable sets are calculated for each microgrid and then composited for the overall system dynamics evaluation. Extensive tests in the full paper will demonstrate the influence of communication latency on system dynamics and validate reachable sets can bound a system's all dynamic trajectories subject to latency changes. The test results also offer an insight into designing and managing the communication network for enhancing the system's resilience.

Li, Yan↗

DOE OSTI2026

wa-hls4ml: A Benchmark and Surrogate Models for hls4ml Resource and Latency Estimation

As machine learning (ML) is increasingly implemented in hardware to address real-time challenges in scientific applications, the development of advanced toolchains has significantly reduced the time required to iterate on various designs. These advancements have solved major obstacles, but also exposed new challenges. For example, processes that were not previously considered bottlenecks, such as hardware synthesis, are becoming limiting factors in the rapid iteration of designs. To mitigate these emerging constraints, multiple efforts have been undertaken to develop an ML-based surrogate model that estimates resource usage of ML accelerator architectures. We introduce wa-hls4ml, a benchmark for ML accelerator resource and latency estimation, and its corresponding initial dataset of over 680,000 fully connected and convolutional neural networks, all synthesized using hls4ml and targeting Xilinx FPGAs. The benchmark evaluates the performance of resource and latency predictors against several common ML model architectures, primarily originating from scientific domains, as exemplar models, and the average performance across a subset of the dataset. Additionally, we introduce GNN- and transformer-based surrogate models that predict latency and resources for ML accelerators. We present the architecture and performance of the models and find that the models generally predict latency and resources for the 75% percentile within several percent of the synthesized resources on the synthetic test dataset.

Hawks, Benjamin [Fermilab] (ORCID:0000000157000288↗

DOE OSTI2025

wa-hls4ml: A Benchmark and Surrogate Models for hls4ml Resource and Latency Estimation

As machine learning (ML) is increasingly implemented in hardware to address real-time challenges in scientific applications, the development of advanced toolchains has significantly reduced the time required to iterate on various designs. These advancements have solved major obstacles, but also exposed new challenges. For example, processes that were not previously considered bottlenecks, such as hardware synthesis, are becoming limiting factors in the rapid iteration of designs. To mitigate these emerging constraints, multipleefforts have been undertaken to develop an ML-based surrogate model that estimates resource usage of synthesized ML accelerator architectures. We introduce wa-hls4ml, a benchmark for ML accelerator resource and latency estimation, and its corresponding initial dataset of over 680 000 fully connected and convolutional neural networks, all synthesized using hls4ml and targeting Xilinx FPGAs. The benchmark evaluates the performance of resource and latency predictors against several common ML model architectures, primarily originating from scientific domains, as exemplar models, and the average performance across a subset of the dataset. Additionally, we introduce GNN- and transformer-based surrogate models that predict latency and resources for ML accelerators. We present the architecture and performance of the models and find that the models generally predict latency and resources for the 75% percentile within several percent of the synthesized resources on the synthetic test dataset.

Hawks, Benjamin G. [Fermilab]↗

NASA NTRS1993

Using Transparent Informed Prefetching (TIP) to reduce file read latency

As processor performance gains continue to outstrip Input/Output gains, I/O performance is becoming critical to overall system performance. File read latency is the most significant bottleneck for high performance I/O. Other aspects of I/O performance benefit from recent advances in disk bandwidth and throughput resulting from disk arrays, and in write performance derived from buffered write behind and the Log-structured File System. The access gap problem limiting improvements in read latency is exacerbated by distributed file systems operating over networks with diverse bandwidth. Focus is on extending the power of caching and prefetching to reduce file read latencies by exploiting hints from high-levels of a system. Such Transparent Informed Prefetching, TIP, and its benefits are described. It is argued that hints that disclose high level knowledge are a means for transferring optimization information across, without violating, module boundaries. How TIP can be used to convert the high throughput of new technologies such as disk arrays and log-structured file systems into low latency for applications is discussed. Our preliminary experiments show reductions in wall - clock execution time of 13 percent and 20 percent for a multiple module compilation tool (make) accessing data on a local disk and remote Coda file server, respectively, and a reduction of 30 percent for a text search (grep) remotely accessing many small files.

Patterson, R. H.↗