Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “workload”

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 343 records · Page 19

Re-Evaluating Virtual Reality Manipulation Techniques for Precise Alignment of Complex 3D Objects

Prior research has developed a number of manipulation techniques that can achieve precise object placement in virtual reality, but studies of these techniques typically use simple objects. We conducted a study comparing two existing techniques, (AMP-IT and WISDOM), during alignment of objects with complex geometry to evaluate the potential influence of geometric complexity on performance, usability, workload and preference. Our findings indicate that participants had faster completion times and higher trial completion rates with AMP-IT on high-precision alignment tasks, contrary to earlier findings that used simple objects. Yet WISDOM is still preferred and considered more usable, despite increased workload and poorer performance, exposing participants' willingness to trade objective performance for comfort during use.

97 MATHEMATICS AND COMPUTING↗

Automated data processing architecture for the Gemini Planet Imager Exoplanet Survey

The Gemini Planet Imager Exoplanet Survey (GPIES) is a multiyear direct imaging survey of 600 stars to discover and characterize young Jovian exoplanets and their environments. We have developed an automated data architecture to process and index all data related to the survey uniformly. An automated and flexible data processing framework, which we term the Data Cruncher, combines multiple data reduction pipelines (DRPs) together to process all spectroscopic, polarimetric, and calibration data taken with GPIES. With no human intervention, fully reduced and calibrated data products are available less than an hour after the data are taken to expedite follow up on potential objects of interest. The Data Cruncher can run on a supercomputer to reprocess all GPIES data in a single day as improvements are made to our DRPs. A backend MySQL database indexes all files, which are synced to the cloud, and a front-end web server allows for easy browsing of all files associated with GPIES. To help observers, quicklook displays show reduced data as they are processed in real time, and chatbots on Slack post observing information as well as reduced data products. Together, the GPIES automated data processing architecture reduces our workload, provides real-time data reduction, optimizes our observing strategy, and maintains a homogeneously reduced dataset to study planet occurrence and instrument performance.

79 ASTRONOMY AND ASTROPHYSICS↗

Software-controlled, highly automated intrafraction prostate motion correction with intrafraction stereographic targeting: System description and clinical results

Purpose: A new system for software-controlled, highly automated correction of intrafraction prostate motion,'' intrafraction stereographic targeting'' (iSGT), is described and evaluated. Methods: At our institute, daily prostate positioning is routinely performed at the start of treatment beam using stereographic targeting (SGT). iSGT was implemented by extension of the SGT software to facilitate fast and accurate intrafraction motion corrections with minimal user interaction. iSGT entails megavoltage (MV) image acquisitions with the first segment of selected IMRT beams, automatic registration of implanted markers, followed by remote couch repositioning to correct for intrafraction motion above a predefined threshold, prior to delivery of the remaining segments. For a group of 120 patients, iSGT with corrections for two nearly lateral beams was evaluated in terms of workload and impact on effective intrafraction displacements in the sagittal plane. Results: SDs of systematic ({Sigma}) and random ({sigma}) displacements relative to the planning CT measured directly after initial SGT setup correction were <0.5 and <0.8 mm, respectively. Without iSGT corrections, effective {Sigma} and {sigma} for the 11-min treatments would increase to {Sigma}{sub eff} < 1.1 mm and {sigma}{sub eff} < 1.2 mm. With the iSGT procedure with an action level of 4 mm, effective positioning errors were reduced to {Sigma}{sub eff} < 0.8 mm and {sigma}{sub eff} < 1.0 mm, with 23.1% of all fractions requiring a correction. Computer simulations demonstrated that with an action level of 2 mm, the errors would have been reduced to {Sigma}{sub eff} < 0.6 mm and {sigma}{sub eff} < 0.7 mm, requiring corrections in 82.4% of the fractions. Because iSGT is highly automated, the extra time added by iSGT is <30 s if a correction is required. Conclusions: Without increasing imaging dose, iSGT successfully reduces intrafraction prostate motion with minimal workload and increase in fraction time. An action level of 2 mm is recommended.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

A study of respiration-correlated cone-beam CT scans to correct target positioning errors in radiotherapy of thoracic cancer

Purpose: There is increasingly widespread usage of cone-beam CT (CBCT) for guiding radiation treatment in advanced-stage lung tumors, but difficulties associated with daily CBCT in conventionally fractionated treatments include imaging dose to the patient, increased workload and longer treatment times. Respiration-correlated cone-beam CT (RC-CBCT) can improve localization accuracy in mobile lung tumors, but further increases the time and workload for conventionally fractionated treatments. This study investigates whether RC-CBCT-guided correction of systematic tumor deviations in standard fractionated lung tumor radiation treatments is more effective than 2D image-based correction of skeletal deviations alone. A second study goal compares respiration-correlated vs respiration-averaged images for determining tumor deviations. Methods: Eleven stage II-IV nonsmall cell lung cancer patients are enrolled in an IRB-approved prospective off-line protocol using RC-CBCT guidance to correct for systematic errors in GTV position. Patients receive a respiration-correlated planning CT (RCCT) at simulation, daily kilovoltage RC-CBCT scans during the first week of treatment and weekly scans thereafter. Four types of correction methods are compared: (1) systematic error in gross tumor volume (GTV) position, (2) systematic error in skeletal anatomy, (3) daily skeletal corrections, and (4) weekly skeletal corrections. The comparison is in terms of weighted average of the residual GTV deviations measured from the RC-CBCT scans and representing the estimated residual deviation over the treatment course. In the second study goal, GTV deviations computed from matching RCCT and RC-CBCT are compared to deviations computed from matching respiration-averaged images consisting of a CBCT reconstructed using all projections and an average-intensity-projection CT computed from the RCCT. Results: Of the eleven patients in the GTV-based systematic correction protocol, two required no correction, seven required a single correction, one required two corrections, and one required three corrections. Mean residual GTV deviation (3D distance) following GTV-based systematic correction (mean {+-} 1 standard deviation 4.8 {+-} 1.5 mm) is significantly lower than for systematic skeletal-based (6.5 {+-} 2.9 mm, p= 0.015), and weekly skeletal-based correction (7.2 {+-} 3.0 mm, p= 0.001), but is not significantly lower than daily skeletal-based correction (5.4 {+-} 2.6 mm, p= 0.34). In two cases, first-day CBCT images reveal tumor changes-one showing tumor growth, the other showing large tumor displacement-that are not readily observed in radiographs. Differences in computed GTV deviations between respiration-correlated and respiration-averaged images are 0.2 {+-} 1.8 mm in the superior-inferior direction and are of similar magnitude in the other directions. Conclusions: An off-line protocol to correct GTV-based systematic error in locally advanced lung tumor cases can be effective at reducing tumor deviations, although the findings need confirmation with larger patient statistics. In some cases, a single cone-beam CT can be useful for assessing tumor changes early in treatment, if more than a few days elapse between simulation and the start of treatment. Tumor deviations measured with respiration-averaged CT and CBCT images are consistent with those measured with respiration-correlated images; the respiration-averaged method is more easily implemented in the clinic.

60 APPLIED LIFE SCIENCES↗

SU-E-T-90: Concrete Forward-Scatter Fractions for Radiotherapy Shielding Applications

Purpose: There is little instruction within the primary shielding guidance document NCRP 151 for vault designs where the primary beam intercepts the maze. We have conducted a Monte-Carlo study to characterize forward-scattered radiation from concrete barriers with the intent of quantifying what amount of additional shielding outside the primary beam is needed in this situation. Methods: We reproduced our vault in MCNP 5 and simulated spectra obtained from the literature and from our treatment planning system for 10 and 18 MV beams. Neutron and gamma-capture contributions were not simulated. Energy deposited was scored at isocenter in a water phantom, within various cells that comprised the maze, and within cells that comprised the vault door. Tracks were flagged that scattered from within the maze to the door and their contributions were tallied separately. Three different concrete mixtures found in the literature were simulated. An empirically derived analytic equation was used for comparison, utilizing patient scatter fractions to approximate the scatter from concrete. Results: Our simulated data confirms that maze-scattered radiation is a significant contribution to total photon dose at the door. It contributes between 20-35% of the photon shielding workload. Forward-scatter fractions for concrete were somewhat dependent on concrete composition and the relative abundance of higher-Z elements. Scatter fractions were relatively insensitive to changes in the primary photon spectrum. Analytic results were of the same magnitude as simulated results. Conclusions: Forward-scattered radiation from the maze barrier needs to be included in the photon workload for shielding calculations in non-standard vault designs. Scatter fractions will vary with concrete composition, but should be insensitive to spectral changes between machine manufacturers. Further plans for investigation include refined scatter fractions for various concrete compositions, scatter fraction measurement, and simulation of photo-nuclear interaction within the maze barrier for high-energy beams.

07 ISOTOPE AND RADIATION SOURCES↗

Automatic learning-based beam angle selection for thoracic IMRT

Purpose: The treatment of thoracic cancer using external beam radiation requires an optimal selection of the radiation beam directions to ensure effective coverage of the target volume and to avoid unnecessary treatment of normal healthy tissues. Intensity modulated radiation therapy (IMRT) planning is a lengthy process, which requires the planner to iterate between choosing beam angles, specifying dose–volume objectives and executing IMRT optimization. In thorax treatment planning, where there are no class solutions for beam placement, beam angle selection is performed manually, based on the planner’s clinical experience. The purpose of this work is to propose and study a computationally efficient framework that utilizes machine learning to automatically select treatment beam angles. Such a framework may be helpful for reducing the overall planning workload. Methods: The authors introduce an automated beam selection method, based on learning the relationships between beam angles and anatomical features. Using a large set of clinically approved IMRT plans, a random forest regression algorithm is trained to map a multitude of anatomical features into an individual beam score. An optimization scheme is then built to select and adjust the beam angles, considering the learned interbeam dependencies. The validity and quality of the automatically selected beams evaluated using the manually selected beams from the corresponding clinical plans as the ground truth. Results: The analysis included 149 clinically approved thoracic IMRT plans. For a randomly selected test subset of 27 plans, IMRT plans were generated using automatically selected beams and compared to the clinical plans. The comparison of the predicted and the clinical beam angles demonstrated a good average correspondence between the two (angular distance 16.8° ± 10°, correlation 0.75 ± 0.2). The dose distributions of the semiautomatic and clinical plans were equivalent in terms of primary target volume coverage and organ at risk sparing and were superior over plans produced with fixed sets of common beam angles. The great majority of the automatic plans (93%) were approved as clinically acceptable by three radiation therapy specialists. Conclusions: The results demonstrated the feasibility of utilizing a learning-based approach for automatic selection of beam angles in thoracic IMRT planning. The proposed method may assist in reducing the manual planning workload, while sustaining plan quality.

60 APPLIED LIFE SCIENCES↗

SU-E-T-670: Radiotherapy Vault Shielding Evaluation Method for a Flattening Filter-Free (FFF) Linac-Practical Considerations and Recommendations

Purpose: To date, there isn’t formal approach for flattening filter-free (FFF) linac vault shielding evaluation, thus, we propose an extension to NCRP#151 to accommodate the recent large number of FFF linac installations.Methods and Materials: We extended the approach in NCRP#151 to design two Truebeam vaults in our new cancer center for hypofractionated treatments. Monte Carlo calculations have characterized primary, scattered, leakage and neutron radiations from FFF-modes. These calculations have shown that: a) FFF primary beam is softer on the central-axis compared to flattening filtered (FF), b) the lateral dose profile is peaked on the central axis and less integral target current is required to generate the same tumor dose with the FF beam. Thus, the TVLs for FFF mode are smaller than those of the FF mode and the scatter functions of the FF mode (NCRP#151) may not be appropriate for FFF-mode, c) the neutron source strength and fluence for 18X-FFF is smaller than 18X-FF, but it is not of a concern here, no 18X-FFF-mode is available on the linac under investigation. Results: These barrier thickness are smaller (12% reduction on the average) than those computed for conventional FF mode with same realistic primary workload since, the primary TVLs used here are smaller and the WL is smaller than the conventional (almost half reduced), keeping the TADR in tolerance. Conclusions: A comprehensive method for shielding barrier calculations based on dedicated data for FFF-mode linacs is highly desired. Meanwhile, we provide an extension to NCRP#151 to accommodate the shielding design of such installations. It is also shown that if a vault is already designed for IMRT/VMAT and SABR hypofractionated treatments with FFF-mode linac, the vault can also be used for a FFF mode linac replacement, leaving some leeway for slightly higher workload on the FFF linac.

61 RADIATION PROTECTION AND DOSIMETRY↗

MO-F-CAMPUS-I-02: Occupational Conceptus Doses From Fluoroscopically-Guided Interventional Procedures

Purpose: The aim of this method was to provide dosimetric data on conceptus dose for the pregnant employee who participates in fluoroscopically-guided interventional procedures. Methods: Scattered air-kerma dose rates were obtained for 17 fluoroscopic projections involved in interventional procedures. These projections were simulated on an anthropomorphic phantom placed on the examination table supine. The operating theater was divided into two grids relative to the long table sides. Each grid consisted of 33 cells spaced 0.50 m apart. During the simulated exposures, at each cell, scatter air-kerma rate was measured at 110 cm from the floor i.e. at the height of the waist of the pregnant worker. Air-kerma rates were divided by the dose area product (DAP) rate of each exposure to obtain normalized data. For each projection, measurements were performed for 3 kVp and 3 filtration values i.e. for 9 different x-ray spectra. All measurements were performed by using a modern C-arm angiographic system (Siemens Axiom Artis, Siemens, Germany) and a radiation meter equipped with an ionization chamber. Results: The results consist of 153 iso-dose maps, which show the spatial distribution of DAP-normalized scattered air-kerma doses at the waist level of a pregnant worker. Conceptus dose estimation is possible using air-kerma to embryo/fetal dose conversion coefficients published in a previous study (J Cardiovasc Electrophysiol, Vol. 16, pp. 1–8, July 2005). Using these maps, occupationally exposed pregnant personnel may select a working position for a certain projection that keeps abdominal dose as low as reasonably achievable. Taking into consideration the regulatory conceptus dose limit for occupational exposure, determination of the maximum workload allowed for the pregnant personnel is also possible. Conclusion: Data produced in this work allow for the anticipation of conceptus dose and the determination of the maximum workload for a pregnant worker from any fluoroscopically-guided interventional procedure. This study was supported by the Greek Ministry of Education and Religious Affairs, General Secretariat for Research and Technology, Operational Program ‘Education and Lifelong Learning’, ARISTIA (Research project: CONCERT)

60 APPLIED LIFE SCIENCES↗

Sci-Thur PM – Brachytherapy 04: Commissioning and Implementation of a Cobalt-60 High Dose Rate Brachytherapy Source

An Eckert & Ziegler Bebig Co0.A86 cobalt 60 high dose rate (HDR) brachytherapy source was commissioned for clinical use. Long-lived Co-60 HDR sources offer potential logistical and economic advantages over Ir-192 sources, and should be considered for low to medium workload brachytherapy departments where modest increases in treatment times are not a factor. In optimized plans, the Co-60 source provides a similar dose distribution to Ir-192 despite the difference in radiation energy. By switching to Co-60, source exchange frequency can be reduced by a factor of 20, resulting in overall financial savings of more than 50% compared to Ir-192 sources. In addition, a reduction in Physicist QA workload of roughly 200 hours over the 5 year life of the Co-60 source is also expected. These benefits should be considered against the modest increases in average treatment time compared to those of Ir-192 sources, as well as the centre-specific needs for operating room shielding modification.

60 APPLIED LIFE SCIENCES↗

iDDS: intelligent distributed dispatch and scheduling for workflow orchestration

The intelligent distributed dispatch and scheduling (iDDS) service is a versatile workflow orchestration system designed for large-scale, distributed scientific computing. iDDS extends traditional workload and data management by integrating data-aware execution, conditional logic, and programmable workflows, enabling automation of complex and dynamic processing pipelines. Originally developed for the ATLAS experiment at the large hadron collider, iDDS has evolved into an experiment-agnostic platform that supports both template-driven workflows and a Function-as-a-Task model for Python-based orchestration. This paper presents the architecture and core components of iDDS, highlighting its scalability, modular message-driven design, and integration with systems such as PanDA and Rucio. We demonstrate its versatility through real-world use cases: fine-grained tape resource optimization for ATLAS, orchestration of large Directed Acyclic Graph (DAG) workflows for the Rubin Observatory, distributed hyperparameter optimization for machine learning applications, active learning for physics analyses, and AI-assisted detector design at the electron–ion collider. By unifying workload scheduling, data movement, and adaptive decision-making, iDDS reduces operational overhead and enables reproducible, high-throughput workflows across heterogeneous infrastructures. We conclude with current challenges and future directions, including interactive, cloud-native, and serverless workflow support.

97 MATHEMATICS AND COMPUTING↗

Making Data Prefetch Smarter: Adaptive Prefetching on POWER7

Hardware data prefetch engines are integral parts of many general purpose server-class microprocessors in the eld to- day. Some prefetch engines allow the user to change some of their parameters. The prefetcher, however, is usually enabled in a default conguration during system bring-up and dynamic reconguration of the prefetch engine is not an autonomic feature of current machines. Conceptually, however, it is easy to infer that commonly used prefetch algorithms, when applied in a xed mode will not help per- formance in many cases. In fact, they may actually de- grade performance due to useless bus bandwidth consump- tion and cache pollution. In this paper, we present an adap- tive prefetch scheme that dynamically modies the prefetch settings in order to adapt to the workload requirements. We implement and evaluate adaptive prefetching in the con- text of an existing, commercial processor, namely the IBM POWER7. Our adaptive prefetch mechanism improves per- formance with respect to the default prefetch setting up to 2.7X and 30% for single-threaded and multiprogrammed workloads, respectively. Categories

Jimenez, Victor↗

Efficient Parallel Discrete Event Simulation on Cloud/Virtual Machine Platforms

Cloud and Virtual machine (VM) technologies present new challenges with respect to performance and monetary cost in executing parallel discrete event simulation (PDES) applications. Due to the introduction of overall cost as a metric, the traditional use of the highest-end computing configuration is no longer the most obvious choice. Moreover, the unique runtime dynamics and configuration choices of Cloud and VM platforms introduce new design considerations and runtime characteristics specific to PDES over Cloud/VMs. Here, an empirical study is presented to guide an understanding of the dynamics, trends, and trade-offs in executing PDES on Cloud/VM platforms. Performance and cost measures obtained from multiple PDES applications executed on the Amazon EC2 Cloud and on a high-end VM host machine reveal new, counterintuitive VM--PDES dynamics and guidelines. One of the critical aspects uncovered is the fundamental mismatch in hypervisor scheduler policies designed for general cloud workloads versus the virtual time ordering needed for PDES workloads. This insight is supported by experimental data revealing the gross deterioration in PDES performance traceable to VM scheduling policy. To overcome this fundamental problem, the design and implementation of a new deadlock-free scheduler algorithm are presented, optimized specifically for PDES applications on VMs. The scalability of our scheduler has been tested up to 128 VMs multiplexed on 32 cores, showing significant improvement in the runtime relative to the default Cloud/VM scheduler. The observations, algorithmic design, and results are timely for emerging cloud/VM-based installations, highlighting the need for PDES-specific support in high performance discrete event simulations on Cloud/VM platforms.

97 MATHEMATICS AND COMPUTING↗

Static and Dynamic Frequency Scaling on Multicore CPUs

Dynamic voltage and frequency scaling (DVFS) adapts CPU power consumption by modifying a processor’s operating frequency (and the associated voltage). Typical approaches employing DVFS involve default strategies such as running at the lowest or the highest frequency, or observing the CPU’s runtime behavior and dynamically adapting the voltage/frequency configuration based on CPU usage. In this paper, we argue that many previous approaches suffer from inherent limitations, such as not account- ing for processor-specific impact of frequency changes on energy for different workload types. We first propose a lightweight runtime-based approach to automatically adapt the frequency based on the CPU workload, that is agnostic of the processor characteristics. We then show that further improvements can be achieved for affine kernels in the application, using a compile-time characterization instead of run-time monitoring to select the frequency and number of CPU cores to use. Our framework relies on a one-time energy characterization of CPU-specific DVFS profiles followed by a compile-time categorization of loop-based code segments in the application. These are combined to determine a priori of the frequency and the number of cores to use to execute the application so as to optimize energy or energy-delay product, outperforming runtime approach. Extensive evaluation on 60 benchmarks and five multi-core CPUs show that our approach systematically outperforms the powersave Linux governor, while improving overall performance.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Modeling Large-Scale Slim Fly Networks Using Parallel Discrete-Event Simulation

As supercomputers approach exascale performance, the increased number of processors translates to an increased demand on the underlying network interconnect. We present that the slim fly network topology, a new low-diameter, low-latency, and low-cost interconnection network, is gaining interest as one possible solution for next-generation supercomputing interconnect systems. In this article, we present a high-fidelity slim fly packet-level model leveraging the Rensselaer Optimistic Simulation System (ROSS) and Co-Design of Exascale Storage (CODES) frameworks. We validate the model with published work before scaling the network size up to an unprecedented 1 million compute nodes and confirming that the slim fly observes peak network throughput at extreme scale. In addition to synthetic workloads, we evaluate large-scale slim fly models with real communication workloads from applications in the Design Forward program with over 110,000 MPI processes. We show strong scaling of the slim fly model on an Intel cluster achieving a peak network packet transfer rate of 2.3 million packets per second and processing over 7 billion discrete events using 128 MPI tasks. Enabled by the strong performance capabilities of the model, we perform a detailed application trace and routing protocol performance study. Lastly, through analysis of metrics such as packet latency, hop count, and congestion, we find that the slim fly network is able to leverage simple minimal routing and achieve the same performance as more complex adaptive routing for tested DOE benchmark applications.

97 MATHEMATICS AND COMPUTING↗

LADR: low-cost application-level detector for reducing silent output corruptions

Applications running on future high performance computing (HPC) systems are more likely to experience transient faults due to technology scaling trends with respect to higher circuit density, smaller transistor size and near-threshold voltage (NTV) operations. A transient fault could corrupt application state without warning, possibly leading to incorrect application output. Such errors are called silent data corruptions (SDCs). In this paper, we present LADR, a low-cost application-level SDC detector for scientific applications. LADR protects scientific applications from SDCs by watching for data anomalies in their state variables (those of scientific interest). It employs compile-time data-flow analysis to minimize the number of monitored variables, thereby reducing runtime and memory overheads while maintaining a high level of fault coverage with low false positive rates. We evaluated LADR with 4 scientific workloads and results show that LADR achieved < 80% fault coverage with only ∼ 3% runtime overheads and ∼ 1% memory overheads. As compared to prior state-of-the-art anomaly-based detection methods, SDC achieved comparable or improved fault coverage, but reduced runtime overheads by 21% ∼ 75%, and memory overheads by 35% ∼ 55% for the evaluated workloads. We believe that such an approach with low memory and runtime overheads coupled with attractive detection precision makes LADR a viable approach for assuring the correct output from large-scale high performance simulations.

Chen, Chao↗

The Minos Computing Library: Efficient Parallel Programming for Extremely Heterogeneous Systems

Hardware specialization has become the silver bullet to achieve efficient high performance, from Systems-on-Chip systems, where hardware specialization can be ``extreme'', to large-scale HPC systems. As the complexity of the systems increases, so does the complexity of programming such architectures in a portable way. This work introduces the Minos Computing Library (MCL), as system software, programming model, and programming model runtime that facilitate programming extremely heterogeneous systems. MCL supports the execution of several multi-threaded applications within the same compute node, performs asynchronous execution of application tasks, efficiently balances computation across hardware resources, and provides performance portability. We show that code developed on a personal desktop automatically scales up to fully utilize powerful workstations with 8 GPUs and down to power-efficient embedded systems. MCL provides up to 17.5x speedup over OpenCL on NVIDIA DGX-1 systems and up to 1.88x speedup on single-GPU systems. In multi-application workloads, MCL dynamically resource allocation provides up to 2.43x performance improvement over manual, static allocation of computing resources.

Gioiosa, Roberto↗

DFSynthesizer: Dataflow-based Synthesis of Spiking Neural Networks to Neuromorphic Hardware

Spiking Neural Networks (SNNs) are an emerging computation model that uses event-driven activation and bio-inspired learning algorithms. SNN-based machine learning programs are typically executed on tile-based neuromorphic hardware platforms, where each tile consists of a computation unit called a crossbar, which maps neurons and synapses of the program. However, synthesizing such programs on an off-the-shelf neuromorphic hardware is challenging. This is because of the inherent resource and latency limitations of the hardware, which impact both model performance, e.g., accuracy, and hardware performance, e.g., throughput. We propose DFSynthesizer, an end-to-end framework for synthesizing SNN-based machine learning programs to neuromorphic hardware. The proposed framework works in four steps. First, it analyzes a machine learning program and generates SNN workload using representative data. Second, it partitions the SNN workload and generates clusters that fit on crossbars of the target neuromorphic hardware. Third, it exploits the rich semantics of the Synchronous Dataflow Graph (SDFG) to represent a clustered SNN program, allowing for performance analysis in terms of key hardware constraints such as number of crossbars, dimension of each crossbar, buffer space on tiles, and tile communication bandwidth. Finally, it uses a novel scheduling algorithm to execute clusters on crossbars of the hardware, guaranteeing hardware performance. We evaluate DFSynthesizer with 10 commonly used machine learning programs. Our results demonstrate that DFSynthesizer provides a much tighter performance guarantee compared to current mapping approaches.

Computer Science↗

Exploring the Use of Novel Spatial Accelerators in Scientific Applications

Driven by the need to find alternative accelerators which can viably replace GPUs in next-generation Supercomputing systems, this paper proposes a methodology to enable agile application/hardware co-design. The application-first methodology provides the ability to come up with design of accelerators while working with real-world workloads, available accelerators, and system software. The iterative design process targets a set of kernels in a workload for performance estimates that can prune the design space for later phases of detailed architectural evaluations. To this effect, in this paper, a novel data-parallel device model is introduced that simulates the latency of performance-sensitive operations in an accelerator including data transfers and kernel computation using multi-core CPUs. The use of off-the-shelf simulators, such as pre-RTL simulator Aladdin or multiple tools available for exploring the design of deep neural network accelerators (e.g., Timeloop) is demonstrated for evaluation of various accelerator designs using applications with realistic inputs. Examples of multiple device configurations that are instantiable in a system are explored to evaluate the performance benefit of deploying novel accelerators. The proposed device is integrated with a programming model and system software to potentially explore the impacts of high-level programming languages/compilers and low-level effects such as task scheduling on multiple accelerators. We analyze our methodology for a set of applications that represent high-performance computing (HPC) and graph analytics. The applications include a computational chemistry kernel realized using tensor contractions, triangle counting, GraphSAGE and Breadth-first Search. These applications include kernels such as dense matrix-dense matrix multiplication, sparse matrix-spare matrix multiplication, and sparse matrix-dense vector multiplication. Our results indicate potential performance benefits and insights for system design by including accelerators that realize these kernels along-side general purpose accelerators.

AI, codesign, Accelerated Computing, Modeling and ↗