Search NASA⌕ Search

SEARCH · Search NASA

Results for “Workload Analysis”

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 163 records · Page 9
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

SoK: What does it Mean to Benchmark Database Forensics?

Relational Database Management Systems are the backbone of modern enterprises and public-sector services, and are thus frequent targets of security incidents, insider threats, and thorough regulatory audits. Consequently, databases have become key sources of digital evidence, requiring investigators to reconstruct past activity from audit logs, transaction logs, and backups. Although benchmarking frameworks such as those developed by the Transaction Processing Performance Council (TPC) are widely used to evaluate database performance, they do not capture forensic requirements such as evidentiary completeness, tamper-evidence, chain of custody, or regulatory compliance under GDPR and CCPA. This survey examines the emerging domain of forensic database benchmarking. We gathered prior research on database forensics, secure logging, and tamper-evident data structures; we analyze modern forensic-ready features in commercial and open-source systems (SQL Server Ledger, Oracle Blockchain Tables, PostgreSQL pgAudit, Db2 Audit, Aurora Database Activity Streams, Oracle Real Application Security and IBM Guardium) and assess why existing benchmarks are insufficient. We propose forensic workloads, metrics, and methodologies that incorporate adversarial stressors, deleted-record recovery, and backup analysis. We also identify open research problems and call for a community-driven forensic benchmark suite. The result is an idea for evaluating not only database performance but also forensic soundness, bridging the gap between system engineering, compliance, and digital investigations.

Lenard, Ben↗
NASA NTRS2006

NASA EVM Overview and Case Study

The presentation gives an overview of the National Aeronautics and Space Administration (NASA) Earned Value Management (EVM) structure. We briefly talk about the current EVM high-level policies within NASA and the EVM governing structure. It touches on the roles and responsibilities of EVM Focal Points within the Agency. We will also discuss the approach that MSFC followed in implementing EVM and better data analysis within the Habitat Holding Racks (HHR) Project. We will address the approach used at the Marshall Space Flight Center (MSFC) to effectively equip and support MSFC projects in applying a sound EVM and data analysis process. In addition, we will show metrics associated with the HHR project before and after the implementation of EVM on the project. We will discuss the monthly report, using sample data, that the project manager used each month to assess the performance of the project. The data received from EVM helped create a solid method for assessing the project s performance. The use of EVM data analysis can be an effective and efficient tool in today s environment with increasing workloads and downsizing workforces. EVM provides project managers with information that can be used in the decision making process.

Kerby, Jerald G.↗
NASA NTRS2002

Shuttle Abort Flight Management (SAFM) - Application Overview

One of the most demanding tasks that must be performed by the Space Shuttle flight crew is the process of determining whether, when and where to abort the vehicle should engine or system failures occur during ascent or entry. Current Shuttle abort procedures involve paging through complicated paper checklists to decide on the type of abort and where to abort. Additional checklists then lead the crew through a series of actions to execute the desired abort. This process is even more difficult and time consuming in the absence of ground communications since the ground flight controllers have the analysis tools and information that is currently not available in the Shuttle cockpit. Crew workload specifically abort procedures will be greatly simplified with the implementation of the Space Shuttle Cockpit Avionics Upgrade (CAU) project. The intent of CAU is to maximize crew situational awareness and reduce flight workload thru enhanced controls and displays, and onboard abort assessment and determination capability. SAFM was developed to help satisfy the CAU objectives by providing the crew with dynamic information about the capability of the vehicle to perform a variety of abort options during ascent and entry. This paper- presents an overview of the SAFM application. As shown in Figure 1, SAFM processes the vehicle navigation state and other guidance information to provide the CAU displays with evaluations of abort options, as well as landing site recommendations. This is accomplished by three main SAFM components: the Sequencer Executive, the Powered Flight Function, and the Glided Flight Function, The Sequencer Executive dispatches the Powered and Glided Flight Functions to evaluate the vehicle's capability to execute the current mission (or current abort), as well as more than IS hypothetical abort options or scenarios. Scenarios are sequenced and evaluated throughout powered and glided flight. Abort scenarios evaluated include Abort to Orbit (ATO), Transatlantic Abort Landing (TAL), East Coast Abort Landing (ECAL) and Return to Launch Site (RTLS). Sequential and simultaneous engine failures are assessed and landing footprint information is provided during actual entry scenarios as well as hypothetical "loss of thrust now" scenarios during ascent.

Hu, Howard↗
NASA NTRS2013

Experiment Description and Results for Arrival Operations Using Interval Management with Spacing to Parallel Dependent Runways (IMSPiDR)

The predicted increase in the number of commercial aircraft operations creates a need for improved operational efficiency. Two areas believed to offer increases in aircraft efficiency are optimized profile descents and dependent parallel runway operations. Using Flight deck Interval Management (FIM) software and procedures during these operations, flight crews can achieve by the runway threshold an interval assigned by air traffic control (ATC) behind the preceding aircraft that maximizes runway throughput while minimizing additional fuel consumption and pilot workload. This document describes an experiment where 24 pilots flew arrivals into the Dallas Fort-Worth terminal environment using one of three simulators at NASA?s Langley Research Center. Results indicate that pilots delivered their aircraft to the runway threshold within +/- 3.5 seconds of their assigned time interval, and reported low workload levels. In general, pilots found the FIM concept, procedures, speeds, and interface acceptable. Analysis of the time error and FIM speed changes as a function of arrival stream position suggest the spacing algorithm generates stable behavior while in the presence of continuous (wind) or impulse (offset) error. Concerns reported included multiple speed changes within a short time period, and an airspeed increase followed shortly by an airspeed decrease.

Baxley, Brian T.↗
DOE OSTI2026

Holistic energy analysis method for thermal management architectures of data centers

Modern high-performance computing (HPC) data centers (DCs), particularly those supporting energy-intensive artificial intelligence (AI) workloads, face escalating thermal management challenges that degrade performance through thermal throttling and drive up cooling power consumption and operational costs. To address this challenge, many have developed a wide variety of thermal management solutions (single-phase, two-phase, direct, indirect, hybrid, and more) which attempt to cool HPC DCs effectively while attempting to minimize overall system power consumption. However, the analysis of these solutions and methods to effectively compare one with another is lacking. Overall power usage effectiveness (PUE) and total-power usage effectiveness (TUE) provide a metric to quantify power consumption but fail to identify components in the system which require further optimization. To address this, we propose a holistic analytical framework – the waterfall diagram (WFD) – which leverages a waterfall chart methodology, offering a comprehensive visualization of both the thermal management system loop and heat flow pathways from individual server components to the outdoor ambient. Use of the WFD enables graphical estimations of power efficiency and cooling performance across each component of a DC cooling system and complements Sankey-style energy flow visualizations by additionally resolving stage-wise temperature changes and incremental TUE contributions. The framework is used in conjunction with simulation-based approaches, to conduct a detailed pressure drop and flow distribution analysis aimed at identifying the optimal coolant distribution architecture for a single-phase direct-to-chip water-cooled DC, which serves as the baseline for subsequent WFD analysis. Among the evaluated architectures, the 3 U modular coolant distribution architecture is found to demonstrate the best performance, considering minimal pressure drop and uniform flow distribution. In addition, TUE is calculated for each cooling loop component based on its associated pressure drop and corresponding pumping power, which are integrated into the WFD. This correlation between TUE and local temperature offers immediate insight into the power efficiency and thermal performance contributions of individual components, facilitating further development and optimization. Examples of WFD applications are presented under varying thermal loads and ambient conditions, demonstrating reasonable cooling strategies. Notably, the 3 U modular architecture maintains a consistent chip case temperature of 85°C, achieving a TUE of 1.016 at ambient temperature of 47°C, and a TUE of 1.026 at ambient temperature of 52°C. The WFD methodology provides an efficient, holistic, and streamlined framework for DC thermal management architecture assessment and enables design optimization which is important for addressing the thermal-fluidic energy challenges of current and next-generation DCs.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
NASA NTRS2006

Implementing EVM Data Analysis Adding Value from a NASA Project Manager's Perspective

Data Analysis is one of the keys to an effective Earned Value Management (EVM) Process. Project Managers (PM) must continually evaluate data in assessing the health of their projects. Good analysis of data can assist PMs in making better decisions in managing projects. To better support our P Ms, National Aeronautics and Space Administration (NASA) - Marshall Space Flight Center (MSFC) recently renewed its emphasis on sound EVM data analysis practices and processes, During this presentation we will discuss the approach that MSFC followed in implementing better data analysis across its Center. We will address our approach to effectively equip and support our projects in applying a sound data analysis process. In addition, the PM for the Space Station Biological Research Project will share her experiences of how effective data analysis can benefit a PM in the decision making process. The PM will discuss how the emphasis on data analysis has helped create a solid method for assessing the project s performance. Using data analysis successfully can be an effective and efficient tool in today s environment with increasing workloads and downsizing workforces

Counts, Stacy↗
NASA NTRSDate not supplied

Usability Evaluation of Fleet Management Interface for High Density Vertiplex Environments

To meet the rising demand for an Advanced Air Mobility (i.e. urban and rural unmanned aircraft systems) ecosystem, NASA Aeronautics Research Mission Directorate (ARMD) is hosting a series of simulations and flight tests under the High Density Vertiplex sub-project (HDV). HDV aims to develop an integrated automation architecture to support terminal area flight operations. The HDV simulations and flight tests address safety, integration, and operational challenges, while integrated systems and software demonstrate design readiness, robustness, and interoperability. During the initial HDV simulation in 2021, a prototype traffic management tool developed by NASA called Fleet Management Interface (FMI) was tested. FMI was designed to introduce an advanced level of human-automation interaction to aid both Ground Control Station Operators (GCSOs) and Fleet Managers (FMs) in remotely managing flights under their ownership. In a human-in-the-loop simulation, a usability study was conducted with FMI to identify optimal approaches for displaying information for human operators using subjective measures of usability, workload, situation awareness, risk, and trust, along with qualitative feedback. This study consisted of task analysis in which GCSO and FM subjects used an Urban Air Mobility (UAM) environment to develop and execute a plan for two different traffic scenarios of remotely controlled vehicles. In each scenario, a controlled vehicle completed a takeoff, active flight, and landing sequence while automated traffic flew in the background at a rate of 20 operations per hour. In the first scenario, the controlled vehicle flew a nominal route with takeoff and landing at the same vertiport. In the second scenario, the controlled vehicle started on the nominal route, then diverted to an unplanned location mid-flight. Results showed that self-reported performance, usability, trust, and situation awareness ratings of FMI were moderately to strongly high. There were small differences between scenarios, with Scenario 2 being perceived as more unstable, complex, variable, risky, and potentially harmful than Scenario 1. Furthermore, participants described improvements that could be made to create a better user experience. For example, users would like greater configurability of the interface based on their personal information requirements, and they would like the opportunity to review routes before assigning them. The results from this study will inform future development of the FMI with the end goal of creating a reference automation tool for airspace management procedures in AAM. The FMI could be introduced as a potential way to reduce dependency on traditional air navigation services through increased automation in high density vertiplex environments.

vertiplex↗
NASA NTRSDate not supplied

Usability Evaluation of Fleet Management Interface for High Density Vertiplex Environments

To meet the rising demand for an Advanced Air Mobility (AAM) (i.e. urban and rural unmanned aircraft systems) ecosystem, the NASA Aeronautics Research Mission Directorate (ARMD) is hosting a series of simulations and flight tests under the High Density Vertiplex sub-project (HDV) to prototype and study the effectiveness AAM capabilities under various operational contexts. HDV aims to develop an integrated automation architecture to support terminal area flight operations. The HDV simulations and flight tests address safety, integration, and operational challenges, while integrated systems and software demonstrate design readiness, robustness, and interoperability. During the initial HDV simulation in 2021, a prototype traffic management tool developed by NASA called the Fleet Management Interface (FMI) was tested. FMI was designed to introduce an advanced level of human-automation interaction to aid both Ground Control Station Operators (GCSOs) and Fleet Managers (FMs) in remotely managing flights under their ownership. In a human-in-the-loop simulation, a usability study was conducted with the FMI to identify optimal approaches for displaying information to human operators using subjective measures of usability, workload, situation awareness, risk, and trust, along with qualitative feedback. This study consisted of task analysis in which GCSO and FM subjects used an Urban Air Mobility (UAM) environment to develop and execute a plan for two different traffic scenarios of remotely controlled vehicles. In each scenario, a remotely controlled vehicle completed a takeoff, active flight, and landing sequence while simulated traffic flew in the background at a rate of 20 operations per hour. In the first scenario, the controlled vehicle flew a nominal route with takeoff and landing at the same vertiport. In the second scenario, the controlled vehicle started on the nominal route, then diverted to an unplanned location mid-flight. Results showed that self- reported performance, usability, trust, and situation awareness ratings of FMI were moderately to strongly high. There were small differences between scenarios, with Scenario 2 being perceived as more unstable, complex, variable, risky, and potentially harmful than Scenario 1. Furthermore, participants described improvements that could be made to create a better user experience. For example, users suggested customizable interfaces to accommodate information display preferences, and the ability to review routes before assigning them. The results from this study will inform future development of the FMI with the end goal of creating a reference automation tool for airspace management procedures in AAM. The FMI could serve to reduce dependency on traditional air navigation services through increased automation in high density vertiplex environments.

Fleet manager↗
NASA NTRS2014

Data Analysis & Statistical Methods for Command File Errors

This paper explains current work on modeling for managing the risk of command file errors. It is focused on analyzing actual data from a JPL spaceflight mission to build models for evaluating and predicting error rates as a function of several key variables. We constructed a rich dataset by considering the number of errors, the number of files radiated, including the number commands and blocks in each file, as well as subjective estimates of workload and operational novelty. We have assessed these data using different curve fitting and distribution fitting techniques, such as multiple regression analysis, and maximum likelihood estimation to see how much of the variability in the error rates can be explained with these. We have also used goodness of fit testing strategies and principal component analysis to further assess our data. Finally, we constructed a model of expected error rates based on the what these statistics bore out as critical drivers to the error rate. This model allows project management to evaluate the error rate against a theoretically expected rate as well as anticipate future error rates.

Correlation Analysis↗
NASA NTRS1993

Brain wave correlates of attentional states: Event related potentials and quantitative EEG analysis during performance of cognitive and perceptual tasks

The increased use of automation in the cockpits of commercial planes has dramatically decreased the workload requirements of pilots, enabling them to function more efficiently and with a higher degree of safety. Unfortunately, advances in technology have led to an unexpected problem: the decreased demands on pilots have increased the probability of inducing 'hazardous states of awareness.' A hazardous state of awareness is defined as a decreased level of alertness or arousal which makes an individual less capable of reacting to unique or emergency types of situations. These states tend to be induced when an individual is not actively processing information. Under such conditions a person is likely to let his/her mind wander, either to internal states or to irrelevant external conditions. As a result, they are less capable of reacting quickly to emergency situations. Since emergencies are relatively rare, and since the high automated cockpit requires progressively decreasing levels of engagement, the probability of being seduced into a lowered state of awareness is increasing. This further decreases the readiness of the pilot to react to unique circumstances such as system failures. The HEM Lab at NASA-Langley Research Center has been studying how these states of awareness are induced and what the physiological correlates of these different states are. Specifically, they have been interested in studying electroencephalographic (EEG) measures of different states of alertness to determine if such states can be identified and, hopefully, avoided. The project worked on this summer involved analyzing the EEG and the event related potentials (ERP) data collected while subjects performed under two conditions. Each condition required subjects to perform a relatively boring vigilance task. The purpose of using these tasks was to induce a decreased state of awareness while still requiring the subject to process information. Each task involved identifying an infrequently presented target stimulus. In addition to the task requirements, irrelevant tones were presented in the background. Research has shown that even though these stimuli are not attended, ERP's to them can still be elicited. The amplitude of the ERP waves has been shown to change as a function of a person's level of alertness. ERP's were also collected and analyzed for the target stimuli for each task. Brain maps were produced based on the ERP voltages for the different stimuli. In addition to the ERP's, a quantitative EEG (QEEG) was performed on the data using a fast Fourier technique to produce a power spectral analysis of the EEG. This analysis was conducted on the continuous EEG while the subjects were performing the tasks. Finally, a QEEG was performed on periods during the task when subjects indicated that they were in an altered state of awareness. During the tasks, subjects were asked to indicate by pressing a button when they realized their level of task awareness had changed. EEG epochs were collected for times just before and just after subjects made this reponse. The purpose of this final analysis was to determine whether or not subjective indices of level of awareness could be correlated with different patterns of EEG.

Freeman, Frederick G.↗
DOE OSTI2022

Basin-Wide Approaches to Hydropower Licensing

This presentation summarizes findings from a 2020 NREL technical report including the FERC relicensing process for non-federal hydropower projects, analyzes FERC-licensed projects with license expiration dates from 2018-2037, discusses FERC's authority to coordinate license terms of hydropower projects within a shared river basin and provides basin-wide case studies and considerations to hydropower relicensing. In the United States there are 1,043 active FERC-licensed hydroelectric projects with a total capacity of 56,097 MW. Of those, 647 with a total capacity of 21,870 MW have license expiration dates from 2018-2037. The expected workload in conjunction with the time, cost, and complexity associated with issuing a single new license has led to initiatives that aim to increase the efficiency of the relicensing process. Federal and state regulators and licensees in California, Maine, New York, and Wisconsin have begun to develop approaches to look at hydropower relicensing as part of a larger system - a river basin. These basin-wide approaches seek to coordinate different stages of the relicensing process for multiple projects at the same time. The goal of these basin-wide approaches is to increase the efficiency of the relicensing process and allow for a more comprehensive analysis of the cumulative impacts of the projects within the basin. These innovative approaches to relicensing could also reduce the workload for regulators and costs associated with relicensing. This presentation discusses some of the findings from the Basin-wide Approaches to Hydropower Relicensing: Case Studies and Considerations including a summary of the key considerations for the FERC relicensing process for non-federal hydropower projects, key findings from the analysis of FERC-licensed projects with license expiration dates from 2018-2037, discusses FERC's authority to coordinate license terms of hydropower projects within a shared river basin and provides basin-wide case studies and considerations to hydropower relicensing.

basin-wide↗
NASA NTRS1990

An in-flight investigation of symbology drive law improvements to an operational attack helicopter

A display drive-law design technique developed at Ames Research Center was applied to the design of the pilot-controlled symbol on a display used in an operational helicopter. A comparison of the newly designed laws and of one documented version of the operational laws was then performed using an in-flight simulation of the helicopter on the NASA/Army CH-47B. Flight-test results indicated that the Ames-developed drive laws were superior to this version of the operational drive laws in terms of task performance and pilot workload. Subsequent postflight-test information suggested that the documentation of the operational drive laws may have been incomplete or in error, even though pilot opinion had corroborated the correctness of the documented operational display in flight. Accordingly, based on additional information, another version of the operational display laws was developed and analyzed, although not flight tested. The revised operational version was compared analytically with the two versions tested in flight. This analysis corroborated that the Ames-developed laws would be superior to this revised version of the operational laws also. A conclusion is that close attention should be paid to the future specification, analysis, and documentation of displayed command-element dynamics, since they have an important effect on task performance and pilot workload.

Schroeder, Jeffery A.↗
NASA NTRS1984

An experimental evaluation of the Sternberg task as a workload metric for helicopter Flight Handling Qualities (FHQ) research

The objective was to determine whether the Sternberg item-recognition task, employed as a secondary task measure of spare mental capacity for flight handling qualities (FHQ) simulation research, could help to differentiate between different flight-control conditions. FHQ evaluations were conducted on the Vertical Motion Simulator at Ames Research Center to investigate different primary flight-control configurations, and selected stability and control augmentation levels for helicopters engaged in low-level flight regimes. The Sternberg task was superimposed upon the primary flight-control task in a balanced experimental design. The results of parametric statistical analysis of Sternberg secondary task data failed to support the continued use of this task as a measure of pilot workload. In addition to the secondary task, subjects provided Cooper-Harper pilot ratings (CHPR) and responded to workload questionnaire. The CHPR data also failed to provide reliable statistical discrimination between FHQ treatment conditions; some insight into the behavior of the secondary task was gained from the workload questionnaire data.

Hemingway, J. C.↗
NASA NTRS1984

The Sternberg Task as a Workload Metric in Flight Handling Qualities Research

The objective of this research was to determine whether the Sternberg item-recognition task, employed as a secondary task measure of spare mental capacity for flight handling qualities (FHQ) simulation research, could help to differentiate between different flight-control conditions. FHQ evaluations were conducted on the Vertical Motion Simulator at Ames Research Center to investigate different primary flight-control configurations, and selected stability and control augmentation levels for helicopers engaged in low-level flight regimes. The Sternberg task was superimposed upon the primary flight-control task in a balanced experimental design. The results of parametric statistical analysis of Sternberg secondary task data failed to support the continued use of this task as a measure of pilot workload. In addition to the secondary task, subjects provided Cooper-Harper pilot ratings (CHPR) and responded to a workload questionnaire. The CHPR data also failed to provide reliable statistical discrimination between FHQ treatment conditions; some insight into the behavior of the secondary task was gained from the workload questionnaire data.

Hemingway, J. C.↗
NASA NTRS1998

Short Haul Civil Tiltrotor Study in MIDAS: Auto versus Manual Nacelle Procedures for Commanded Go-Around

Tiltrotor aircraft combine the speed and range of a turboprop performance with the ability to take off and land in a vertical mode like a helicopter. These aircraft will transport passengers from city center to city center and from satellite airports to major hub airports to make connections to long range travel. The Short Haul Civil Tiltrotor (SH(CT)) being studied by NASA is a concept 40 passenger civil tiltrotor (CTR) transport. The Man-machine Integration Design and Analysis System (MIDAS) was used to evaluate human performance in terms of crew procedures and pilot workload for a simulated 40 passenger Civil Tiltrotor Transport on a steep approach to a vertiport. The scenario for the simulation was a normal approach to the vertiport that is interrupted by a commanded go-around at the landing decision point. The simulation contrasted an automated discrete nacelle mode control with a fully manual nacelle control mode for the go-around. The MIDAS simulation showed that the pilot task loading during approach and for the commanded go-around is high and that pilot workload is near capacity throughout. The go-around in manual nacelle mode was most demanding, resulting in additional time requirements to complete necessary tasks.

Atencio, Adolph, Jr.↗
NASA NTRS2003

Science Goal Driven Automation for NASA Missions: The Science Goal Monitor

Infusion of automation technologies into NASA s future missions will be essential not only to achieve substantial reduction in mission operations staff and costs, but also in order to both effectively handle an exponentially increasing volume of scientific data and to successfully meet dynamic, opportunistic scientific goals and objectives. Current spacecraft operations cannot respond to science driven events, such as intrinsically variable or short-lived phenomena in a timely manner. For such investigations, we must teach our platforms to dynamically understand, recognize, and react to the scientists goals. While much effort has gone into automating routine spacecraft operations to reduce human workload and hence costs, applying intelligent automation to the science side, i.e., science data acquisition, data analysis and reactions to that data analysis in a timely and still scientifically valid manner, has been relatively under-emphasized.

Korathkar, Anuradha↗
NASA NTRS2004

Science Goal Monitor: Science Goal Driven Automation for NASA Missions

Infusion of automation technologies into NASA s future missions will be essential because of the need to: (1) effectively handle an exponentially increasing volume of scientific data, (2) successfully meet dynamic, opportunistic scientific goals and objectives, and (3) substantially reduce mission operations staff and costs. While much effort has gone into automating routine spacecraft operations to reduce human workload and hence costs, applying intelligent automation to the science side, i.e., science data acquisition, data analysis and reactions to that data analysis in a timely and still scientifically valid manner, has been relatively under-emphasized. In order to introduce science driven automation in missions, we must be able to: capture and interpret the science goals of observing programs, represent those goals in machine interpretable language; and allow spacecrafts onboard systems to autonomously react to the scientist's goals. In short, we must teach our platforms to dynamically understand, recognize, and react to the scientists goals. The Science Goal Monitor (SGM) project at NASA Goddard Space Flight Center is a prototype software tool being developed to determine the best strategies for implementing science goal driven automation in missions. The tools being developed in SGM improve the ability to monitor and react to the changing status of scientific events. The SGM system enables scientists to specify what to look for and how to react in descriptive rather than technical terms. The system monitors streams of science data to identify occurrences of key events previously specified by the scientist. When an event occurs, the system autonomously coordinates the execution of the scientist s desired reactions. Through SGM, we will improve om understanding about the capabilities needed onboard for success, develop metrics to understand the potential increase in science returns, and develop an operational prototype so that the perceived risks associated with increased use of automation can be reduced.

Koratkar, Anuradha↗
DOE OSTI2020

CSB-RNN: A Faster-Than-Realtime RNN Acceleration Framework with Compressed Structured Blocks

Recurrent Neural Networks (RNN) is widely applied to temporal sequence analysis, where real-time performance is usually in demand. However, RNN suffers a heavy computational workload as the model comes with a large weight matrix. To alleviate the pain, model compression (pruning) schemes have been proposed for RNN that pruning the redundant (near-zero) weight-values. On the one hand, the non-structured pruning methods achieve a considerable pruning rate while bringing the computational irregularity, which is un-friendly to parallel-hardware. On the other hand, the existing structured pruning methods consider the hardware parallelism; However, they suffer a poor pruning rate due to the restrict constraints on pruning structure. This paper presents CSB-RNN, an optimized full-stack RNN framework with the novel compressed structured block (CSB) technique. The CSB-pruned RNN model comes with both fine-granularity that benefits the pruning rate and regular structure that facilitates the hardware-parallelism. Further, we propose a novel hardware architecture for inferencing the CSB-pruned model. Different from conventional parallel hardware, this architecture solves the block-workload imbalance issue and achieves an over 95% hardware utilization. With the experiments on 10 RNN models in 5 application domains, the CSB-RNN realizes 7×-20× lossless compression and up to 50× acceptable lossy-compression, which is 2×-7× to the prior art. With the addition of the novel hardware, the compressed-RNN inference reaches a super real-time latency of 10-400µs with FPGA implementation.

Shi, Runbin↗