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 217 records · Page 12

Coordinating Garbage Collection for Arrays of Solid-state Drives

Although solid-state drives (SSDs) offer significant performance improvements over hard disk drives (HDDs) for a number of workloads, they can exhibit substantial variance in request latency and throughput as a result of garbage collection (GC). When GC conflicts with an I/O stream, the stream can make no forward progress until the GC cycle completes. GC cycles are scheduled by logic internal to the SSD based on several factors such as the pattern, frequency, and volume of write requests. When SSDs are used in a RAID with currently available technology, the lack of coordination of the SSD-local GC cycles amplifies this performance variance. We propose a global garbage collection (GGC) mechanism to improve response times and reduce performance variability for a RAID of SSDs. We include a high-level design of SSD-aware RAID controller and GGC-capable SSD devices and algorithms to coordinate the GGC cycles. We develop reactive and proactive GC coordination algorithms and evaluate their I/O performance and block erase counts for various workloads. Our simulations show that GC coordination by a reactive scheme improves average response time and reduces performance variability for a wide variety of enterprise workloads. For bursty, write-dominated workloads, response time was improved by 69% and performance variability was reduced by 71%. We show that a proactive GC coordination algorithm can further improve the I/O response times by up to 9% and the performance variability by up to 15%. We also observe that it could increase the lifetimes of SSDs with some workloads (e.g. Financial) by reducing the number of block erase counts by up to 79% relative to a reactive algorithm for write-dominant enterprise workloads.

Flash Memory↗

Spatial Locality-Aware Cache Partitioning for Effective Cache Sharing, In: 2015 44th International Conference on Parallel Processing

In modern multi-core processors, last-level caches (LLCs) are typically shared among multiple cores. Previous works have shown that such sharing is beneficial as different workloads have different needs for cache capacity, and logical partitioning of capacity can improve system performance. However, what is missing in previous works on partitioning shared LLCs is that the heterogeneity in spatial locality among workloads has not been explored. In other words, all the cores use the same block/line size in shared LLCs. In this work, we highlight that exploiting spatial locality enables much more effective cache sharing. The fundamental reason is that for many memory intensive workloads, their cache capacity requirements can be drastically reduced when a large block size is employed, therefore they can effectively donate more capacity to other workloads. To leverage spatial locality for cache partitioning effectively, we first propose a simple yet effective mechanism to measure both spatial and temporal locality at run-time. The locality information is then used to determine both the proper block size and the capacity assigned to each workload. Our experiments show that our Spatial Locality-aware Cache Partitioning (SLCP) significantly outperforms the previous works. We also present several case studies that dissect the effectiveness of SLCP compared to the existing approaches.

Computer Science↗

MAPA: Multi-Accelerator Pattern Allocation Policy for Multi-Tenant GPU Servers

Multi-accelerator servers are increasingly being deployed in shared multi-tenant environments (such as in cloud data centers) in order to meet the demands of large-scale compute-intensive workloads. In addition, these accelerators are increasingly being inter-connected in complex topologies and workloads are exhibiting a wider variety of inter-accelerator communication patterns. However, existing al-location policies are ill-suited for these emerging use-cases. Specifically, this work identifies that multi-accelerator workloads are commonly fragmented leading to reduced bandwidth and increased latency for inter-accelerator communication. We propose Multi-Accelerator Pattern Allocation (MAPA), a graph pattern mining approach towards providing generalized allocation support for allocating multi-accelerator workloads on multi-accelerator servers. We demonstrate that MAPA is able to improve the execution time of multi-accelerator workloads and that MAPA is able to provide generalized benefits across various accelerator topologies. Finally, we demonstrate a speedup of 12.4% for75th percentile of jobs with the worst case execution time reduced by up to 35% against baseline policy using MAPA.

Ranganath, Kiran↗

AI4IO: A suite of AI-based tools for IO-aware scheduling

Traditional workload managers do not have the capacity to consider how IO contention can increase job runtime and even cause entire resource allocations to be wasted. Whether from bursts of IO demand or parallel file systems (PFS) performance degradation, IO contention must be identified and addressed to ensure maximum performance. In this paper, we present AI4IO (AI for IO), a suite of tools using AI methods to prevent and mitigate performance losses due to IO contention. AI4IO enables existing workload managers to become IO-aware. Currently, AI4IO consists of two tools: PRIONN and CanarIO. PRIONN predicts IO contention and empowers schedulers to prevent it. CanarIO mitigates the impact of IO contention when it does occur. We measure the effectiveness of AI4IO when integrated into Flux, a next-generation scheduler, for both small- and large-scale IO-intensive job workloads. Our results show that integrating AI4IO into Flux improves the workload makespan up to 6.4%, which can account for more than 18,000 node-h of saved resources per week on a production cluster in our large-scale workload.

Wyatt, II, Michael R.↗

EXPLORING THE POTENTIAL OF SHORT-TIME FOURIER TRANSFORMS FOR ANALYZING SKIN CONDUCTANCE AND PUPILLOMETRY IN REAL-TIME APPLICATIONS

The development of real-time predictors of mental workload is critical for the practical application of augmented cognition to human-machine systems. This paper explores a novel method based on a short-time Fourier transform (STFT) for analyzing galvanic skin conductance (SC) and pupillometry time-series data to extract estimates of mental workload with temporal bandwidth high-enough to be useful for augmented cognition applications. We tested the method in the context of a process control task based on the DURESS simulation developed by Vincente and Pawlak (1994; ported to Java by Cosentino,& Ross, 1999). SC, pupil dilation, blink rate, and visual scanning patterns were measured for four participants actively engaged in controlling the simulation. Fault events were introduced that required participants to diagnose errors and make control adjustments to keep the simulator operating within a target range. We were interested in whether the STFT of these measures would produce visible effects of the increase in mental workload and stress associated with these events. Graphical exploratory data analysis of the STFT showed visible increases in the power spectrum across a range of frequencies directly following fault events. We believe this approach shows potential as a relatively unobtrusive, low-cost, high bandwidth measure of mental workload that could be particularly useful for the application of augmented cognition to human-machine systems.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Human System Simulation in Support of Human Performance Technical Basis at NPPs

This paper focuses on strategies and progress toward establishing the Idaho National Laboratory’s (INL’s) Human Systems Simulator Laboratory at the Center for Advanced Energy Studies (CAES), a consortium of Idaho State Universities. The INL is one of the National Laboratories of the US Department of Energy. One of the first planned applications for the Human Systems Simulator Laboratory is implementation of a dynamic nuclear power plant simulation (NPP) where studies of operator workload, situation awareness, performance and preference will be carried out in simulated control rooms including nuclear power plant control rooms. Simulation offers a means by which to review operational concepts, improve design practices and provide a technical basis for licensing decisions. In preparation for the next generation power plant and current government and industry efforts in support of light water reactor sustainability, human operators will be attached to a suite of physiological measurement instruments and, in combination with traditional Human Factors Measurement techniques, carry out control room tasks in simulated advanced digital and hybrid analog/digital control rooms. The current focus of the Human Systems Simulator Laboratory is building core competence in quantitative and qualitative measurements of situation awareness and workload. Of particular interest is whether introduction of digital systems including automated procedures has the potential to reduce workload and enhance safety while improving situation awareness or whether workload is merely shifted and situation awareness is modified in yet to be determined ways. Data analysis is carried out by engineers and scientists and includes measures of the physical and neurological correlates of human performance. The current approach supports a user-centered design philosophy (see ISO 13407 “Human Centered Design Process for Interactive Systems, 1999) wherein the context for task performance along with the requirements of the end-user are taken into account during the design process and the validity of design is determined through testing of real end users

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Evaluating the potential of disaggregated memory systems for HPC applications

Summary Disaggregated memory is a promising approach that addresses the limitations of traditional memory architectures by enabling memory to be decoupled from compute nodes and shared across a data center. Cloud platforms have deployed such systems to improve overall system memory utilization, but performance can vary across workloads. High‐performance computing (HPC) is crucial in scientific and engineering applications, where HPC machines also face the issue of underutilized memory. As a result, improving system memory utilization while understanding workload performance is essential for HPC operators. Therefore, learning the potential of a disaggregated memory system before deployment is a critical step. This paper proposes a methodology for exploring the design space of a disaggregated memory system. It incorporates key metrics that affect performance on disaggregated memory systems: memory capacity, local and remote memory access ratio, injection bandwidth, and bisection bandwidth, providing an intuitive approach to guide machine configurations based on technology trends and workload characteristics. We apply our methodology to analyze thirteen diverse workloads, including AI training, data analysis, genomics, protein, fusion, atomic nuclei, and traditional HPC bookends. Our methodology demonstrates the ability to comprehend the potential and pitfalls of a disaggregated memory system and provides motivation for machine configurations. Our results show that eleven of our thirteen applications can leverage injection bandwidth disaggregated memory without affecting performance, while one pays a rack bisection bandwidth penalty and two pay the system‐wide bisection bandwidth penalty. In addition, we also show that intra‐rack memory disaggregation would meet the application's memory requirement and provide enough remote memory bandwidth.

Ding, Nan↗

An automated and portable method for selecting an optimal GPU frequency

Power consumption poses a significant challenge in current and emerging graphics processing unit (GPU) enabled high-performance computing systems. In modern GPUs, dynamic voltage frequency scaling (DVFS) appears to be a reliable control to regulate power consumption and performance. However, the DVFS design space is large - hence, brute-force approaches are infeasible to select the optimal frequency. Furthermore, no single frequency can be universally optimal for applications with varying computational intensities. Thus, the application's complexity and the availability of a wide range of frequency settings are a challenge in selecting the optimal frequency configuration for a given GPU workload. To that end, this paper proposes a systematic approach that consists of three steps. The feature characterization study identifies the fine-grain GPU utilization metrics that influence the power consumption and execution time of a given workload. To understand the performance, power, and energy consumption behaviors of a workload across GPU's DVFS design space, we derived analytical power and performance models using the identified fine-grain features. Here, it is shown that the same set of GPU utilization metrics can estimate both the power consumption and execution time while being agnostic of changes to frequency and input sizes. Applying a power control with the single objective of reducing power may cause performance degradation, leading to more energy consumption. A multi-objective approach is proposed to select the optimal GPU DVFS configuration for a workload that reduces power consumption with negligible degradation in performance. The evaluation was conducted using SPEC ACCEL benchmarks and three real applications - NAMD LAMMPS, and LSTM on NVIDIA GV100, GA100, and AMD MI210 GPUs. On average, real applications showed 29.6% energy savings with a performance loss of 5.2% on GA100 and 22.6% energy savings with a performance loss of 4.7% on GV100. Moreover, the proposed models are portable to real applications, GPU architectures, and vendors, and require metric collection at only the default frequency rather than all supported DVFS configurations. Additionally, we conducted a comparison between our models and the GPU assembly instructions (PTX)-based static models. The results revealed a significant reduction in the average error rates, with a decrease from 19.7% to 3.1% for power models and from 29.4% to 5.2% for performance models.

97 MATHEMATICS AND COMPUTING↗

Evaluating adaptive and predictive power management strategies for optimizing visualization performance on supercomputers

Power is becoming an increasingly scarce resource on the next generation of supercomputers, and should be used wisely to improve overall performance. One strategy for improving power usage is hardware overprovisioning, i.e., systems with more nodes than can be run at full power simultaneously without exceeding the system-wide power limit. With this study, we compare two strategies for allocating power throughout an overprovisioned system – adaptation and prediction – in the context of visualization workloads. While adaptation has been suitable for workloads with more regular execution behaviors, it may not be as suitable on visualization workloads, since they can have variable execution behaviors. This study considers a total of 104 experiments, which vary the rendering workload, power budget, allocation strategy, and node concurrency, including tests processing data sets up to 1 billion cells and using up to 18,432 cores across 512 nodes. Overall, we find that prediction is a superior strategy for this use case, improving performance up to 27% compared to an adaptive strategy.

97 MATHEMATICS AND COMPUTING↗

Preparation of the Multi-Site Data Processing at the Vera C. Rubin Observatory

The Vera C. Rubin Observatory’s Legacy Survey of Space and Time (LSST) Camera is scheduled to start taking data in the summer of 2025. The Data Release Production will run the LSST Science Pipe software at data facilities in the US, France and the UK. The LSST Science Pipeline consists of complex directed acyclic graphs (DAGs) of tasks. Rubin will use the Production and Distributed Analysis (PanDA) workflow and workload management system to orchestrate this complex workflow and the distribution of workloads to the data facilities. When run end-to-end by a team of data production staff, this processing (the Science Pipelines, distributed by the workflow and workload management system) is referred to as a 'campaign'. This paper describes the central services and data facility specific services that support this multi-site data process model, including the service deployment infrastructure, the workload and workflow system, the Campaign Management tools, and connection to Rubin Data Management. This paper will also mention the experience of processing the Rubin Commissioning Camera data. All these are part of the effort to scale up the processing capabilities for the expected very large data volume from the LSST Camera.

Yang, Wei [SLAC]↗

Addressing Load Imbalance in Bioinformatics and Biomedical Applications: Efficient Scheduling across Multiple GPUs

Computational bioinformatics and biomedical applications frequently contain heterogeneously sized units of work or tasks, for instance due to variability in the sizes of biological sequences and molecules. Variable-sized workloads lead to load imbalances in parallel implementations which detract from efficiency and performance. Many modern computing resources now have multiple graphics processing units(GPUs) per computer for acceleration. These multiple GPU resources need to be used efficiently through balancing of workloads across the GPUs. OpenMP is a portable directive-based parallel programming API used ubiquitously in bioscience applications to program CPUs; recently, the use of OpenMP directives for GPU acceleration has become possible. Here, motivated by experiences with imbalanced loads in GPU-accelerated bioinformatics applications, we address the load balancing problem using OpenMP task-to-GPU scheduling combined with OpenMP GPU offloading for multiply heterogeneous workloads – loads with both variable input sizes, and simultaneously, variable convergence rates for algorithms with a stochastic component – scheduled across multiple GPUs. We aim to develop strategies which are both easy to use and have lower overheads, and may be incorporated incrementally in existing programs which already make use of OpenMP for CPU-based threading in order to make use of multi-GPU computers. We test different combinations of input size variability and convergence rate variability, and characterize the effects of these different scenarios on the performance of scheduling strategies across multiple GPUs with OpenMP. We present several dynamic scheduling solutions for different parallel patterns, explore optimizations, and provide publicly available example computational kernels to make these strategies easy to use in programs. This work will enable application developers to efficiently and easily use multiple GPUs for imbalanced workloads found in bioinformatics and biomedical applications.

Thavappiragasam, Mathialakan↗

Modeling Cooperative Threads to Project GPU Performance for Adaptive Parallelism

Most accelerators, such as graphics processing units (GPUs) and vector processors, are particularly suitable for accelerating massively parallel workloads. On the other hand, conventional workloads are developed for multi-core parallelism, which often scale to only a few dozen OpenMP threads. When hardware threads significantly outnumber the degree of parallelism in the outer loop, programmers are challenged with efficient hardware utilization. A common solution is to further exploit the parallelism hidden deep in the code structure. Such parallelism is less structured: parallel and sequential loops may be imperfectly nested within each other, neigh boring inner loops may exhibit different concurrency patterns (e.g. Reduction vs. Forall), yet have to be parallelized in the same parallel section. Many input-dependent transformations have to be explored. A programmer often employs a larger group of hardware threads to cooperatively walk through a smaller outer loop partition and adaptively exploit any encountered parallelism. This process is time-consuming and error-prone, yet the risk of gaining little or no performance remains high for such workloads. To reduce risk and guide implementation, we propose a technique to model workloads with limited parallelism that can automatically explore and evaluate transformations involving cooperative threads. Eventually, our framework projects the best achievable performance and the most promising transformations without implementing GPU code or using physical hardware. We envision our technique to be integrated into future compilers or optimization frameworks for autotuning.

Meng, Jiayuan↗

Co-design of Advanced Architectures for Graph Analytics using Machine Learning

A graph is an excellent way of representing relationships among entities. We can use graph analytics to synthesize and analyze such relational data, and extract relevant features that are useful for various tasks such as machine learning. Considering the crucial role of graph analytics in various domains, it is important and timely to investigate the right hardware configurations that can achieve optimal performance for graph workloads on future high-performance computing systems. Design space exploration studies facilitate the selection of appropriate configurations (e.g. memory) to achieve a desired system performance. Recently, the approach of accelerating graph analytics using persistent non-volatile memory has gained a lot of attention. Traditional system simulators such as Gem5 and NVMain can be used to explore the design space of these advanced memory architectures for graph workloads. However, these simulators are slow in execution thus limiting the efficiency of design space exploration studies. To overcome this challenge, we proposed a machine learning based approach to co-design advanced memory architectures for graph workloads. We tested our approach with DRAM, non-volatile memory, and hybrid memory (DRAM+NVM) using a breadth first search benchmark algorithm. Our results showed the applicability of the proposed machine learning based approach to the co-design of the advanced memory architectures. In this paper, we provide recommendations on selecting advanced memory architectures to achieve desired performance for graph workloads. We also discuss the performances of different machine learning models that were considered in this study.

Kurte, Kuldeep↗

Dynamic Data Repartitioning for Load-Balanced Parallel Particle Tracing

We present a novel dynamic load-balancing algorithm based on data repartitioning for parallel particle tracing in flow visualization. Instead of static data assignment, we dynamically repartition the data into blocks and reassign the blocks to processes to balance the workload distribution among the processes. Block repartitioning is performed based on a dynamic workload estimation method that predicts the workload in the flow field on the fly as the input. In our approach, we allow data duplication in the repartitioning, enabling the same data blocks to be assigned to multiple processes. Load balance is achieved by regularly exchanging the blocks (together with the particles in the blocks) among processes according to the output of the data repartitioning. Compared with other load-balancing algorithms, our approach does not need any preprocessing on the raw data and does not require any dedicated process for work scheduling, while it has the capability to balance uneven workload efficiently. Results show improved load balance and high efficiency of our method on tracing particles in both steady and unsteady flow.

data repartitioning↗

Preemptible I/O Scheduling of Garbage Collection for Solid State Drives

Unlike hard disks, flash devices use out-of-place updates operations and require a garbage collection (GC) process to reclaim invalid pages to create free blocks. This GC process is a major cause of performance degradation when running concurrently with other I/O operations as internal bandwidth is consumed to reclaim these invalid pages. The invocation of the GC process is generally governed by a low watermark on free blocks and other internal device metrics that different workloads meet at different intervals. This results in an I/O performance that is highly dependent on workload characteristics. In this paper, we examine the GC process and propose a semipreemptible GC (PGC) scheme that allows GC processing to be preempted while pending I/O requests in the queue are serviced. Moreover, we further enhance flash performance by pipelining internal GC operations and merge them with pending I/O requests whenever possible. Our experimental evaluation of this semi-PGC scheme with realistic workloads demonstrates both improved performance and reduced performance variability. Write-dominant workloads show up to a 66.56% improvement in average response time with a 83.30% reduced variance in response time compared to the non-PGC scheme. In addition, we explore opportunities of a new NAND flash device that supports suspend/resume commands for read, write, and erase operations for fully PGC (F-PGC). Our experiments with an F-PGC enabled flash device show that request response time can be improved by up to 14.57% compared to semi-PGC.

97 MATHEMATICS AND COMPUTING↗

Dynamic Undervolting to Improve Energy Efficiency on Multicore X86 CPUs

Chip manufacturers introduce redundancy at various levels of CPU design to guarantee correct operation, even for worst-case combinations of non-idealities in process variation and system operating conditions. This redundancy is implemented partly in the form of voltage margins. However, for a wide range of real-world execution scenarios these margins are excessive and merely translate to increased power consumption, hindering the effort towards higher-energy efficiency in both HPC and general purpose computing. Our study on the x86-64 Haswell and Skylake multicore microarchitectures reveals-wide voltage margins, which vary across different microarchitectures, different chip parts of the same microarchitecture, and across different workloads. We find that it is necessary to quantify-voltage margins using multi-threaded and multi-instance workloads, as characterization with single-threaded and single-instance workloads that do not stress the CPU to its full capacity typically identifies overly optimistic margins that lead to errors when applied in realistic program execution scenarios. In addition, we introduce, deploy and evaluate a run-time governor that dynamically reduces the supply voltage of modern multicore x86-64 CPUs. Our governor employs a model that takes as input a set of performance metrics which are directly measurable via performance monitoring counters and have high predictive value for the minimum tolerable supply voltage (V min ), to predict and apply the appropriate reduction for the workload at hand. Compared with the conventional DVFS governor, our approach in this study achieves up to 42 percent energy savings for the Skylake family and 34 percent for the Haswell family for complex, real-world applications.

97 MATHEMATICS AND COMPUTING↗

DRAS: Deep Reinforcement Learning for Cluster Scheduling in High Performance Computing

Cluster schedulers are crucial in high-performance computing (HPC). They determine when and which user jobs should be allocated to available system resources. Existing cluster scheduling heuristics are developed by human experts based on their experience with specific HPC systems and workloads. However, the increasing complexity of computing systems and the highly dynamic nature of application workloads have placed tremendous burden on manually designed and tuned scheduling heuristics. More aggressive optimization and automation are needed for cluster scheduling in HPC. In this work, we present an automated HPC scheduling agent named DRAS (Deep Reinforcement Agent for Scheduling) by leveraging deep reinforcement learning. DRAS is built on a hierarchical neural network incorporating special HPC scheduling features such as resource reservation and backfilling. An efficient training strategy is presented to enable DRAS to rapidly learn the target environment. Once being provided a specific scheduling objective given by the system manager, DRAS automatically learns to improve its policy through interaction with the scheduling environment and dynamically adjusts its policy as workload changes. We implement DRAS into a HPC scheduling platform called CQGym. CQGym provides a common platform allowing users to flexibly evaluate DRAS and other scheduling methods such as heuristic and optimization methods. Furthermore, the experiments using CQGym with different production workloads demonstrate that DRAS outperforms the existing heuristic and optimization approaches by up to 50%.

97 MATHEMATICS AND COMPUTING↗

Scatter radiation intensities around a clinical digital breast tomosynthesis unit and the impact on radiation shielding considerations

Purpose: To measure the scattered radiation intensity around a clinical digital breast tomosynthesis (DBT) unit and to provide updated data for radiation shielding design for DBT systems with tungsten-anode x-ray tubes. Methods: The continuous distribution of scattered x-rays from a clinical DBT system (Hologic Selenia Dimensions) was measured within an angular range of 0°–180° using a linear-array x-ray detector (X-Scan 0.8f3-512, Detection Technology, Inc., Finland), which was calibrated for the x-ray spectrum range of the DBT unit. The effects of x-ray field size, phantom size, and x-ray kVp/filter combination were investigated. Following a previously developed methodology by Simpkin, scatter fraction was determined for the DBT system as a function of angle around the phantom center. Detailed calculations of the scatter intensity from a DBT system were demonstrated using the measured scatter fraction data. Results: For the 30 and 35 kVp acquisition, the scatter-to-primary-ratio and scatter fraction data closely matched with data previously measured by Simpkin. However, the measured data from this study demonstrated the nonisotropic distribution of the scattered radiation around a DBT system, with two strong peaks around 25° and 160°. The majority scatter radiation (>70%) originated from the imaging detector assembly, instead of the phantom. With a workload from a previous survey performed at MGH, the scatter air kerma at 1 m from the phantom center for wall/door is 1.76 × 10{sup −2} mGy patient{sup −1}, for floor is 1.64 × 10{sup −1} mGy patient{sup −1}, and for ceiling is 3.66 × 10{sup −2} mGy patient{sup −1}. Conclusions: Comparing to previously measured data for mammographic systems, the scatter air kerma from Holgoic DBT is at least two times higher. The main reasons include the harder primary beam with higher workload (measured with total mAs/week), added tomosynthesis acquisition, and strong small angle forward scattering. Due to the highly conservative initial assumptions, the shielding recommendation from NCRP Report 147 is still sufficient for the Hologic DBT system given the workload from a previous survey at MGH. With the data provided from this study, accurate shielding calculation can be performed for Hologic DBT systems with specific workload and barrier distance.

61 RADIATION PROTECTION AND DOSIMETRY↗