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 235 records · Page 13

A Semi-Preemptive Garbage Collector for Solid State Drives

NAND flash memory is a preferred storage media for various platforms ranging from embedded systems to enterprise-scale systems. Flash devices do not have any mechanical moving parts and provide low-latency access. They also require less power compared to rotating media. Unlike hard disks, flash devices use out-of-update operations and they 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 I/O performance that is highly dependent on workload characteristics. In this paper, we examine the GC process and propose a semi-preemptive GC scheme that can preempt on-going GC processing and service pending I/O requests in the queue. 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-preemptive GC sheme with realistic workloads demonstrate 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-preemptive GC scheme.

99 GENERAL AND MISCELLANEOUS↗

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

Abstract Unlike hard disks, flash devices use out-of-update operations and they 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 I/O performance that is highly dependent on workload characteristics. In this paper, we examine the GC process and propose a semi-preemptible GC 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 semipreemptible GC scheme with realistic workloads demonstrate 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-preemptible GC 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 preemptible GC. Our experiments with a fully preemptible GC enabled flash device show that request response time can be improved by up to 14.57% compared to semi-preemptible GC.

Flash Memory↗

Geomancy: Automated Performance Enhancement through Data Layout Optimization

Large distributed storage systems such as high- performance computing (HPC) systems used by national or international laboratories require sufficient performance and scale for demanding scientific workloads and must handle shifting workloads with ease. Ideally, data is placed in locations to optimize performance, but the size and complexity of large storage systems inhibit rapid effective restructuring of data layouts to maintain performance as workloads shift. To address these issues, we have developed Geomancy, a tool that models the placement of data within a distributed storage system and reacts to drops in performance. Using a combination of machine learning techniques suitable for temporal modeling, Geomancy determines when and where a bottleneck may happen due to changing workloads and suggests changes in the layout that mitigate or prevent them. Our approach to optimizing throughput offers benefits for storage systems such as avoiding potential bottlenecks and increasing overall I/O throughput from 11% to 30%.

Bel, Oceane M.↗

Shape-shifting Elephants: Multi-modal Transport for Integrated Research Infrastructure

Data Acquisition (DAQ) workloads form an important class of scientific network traffic that by its nature (1) flows across different research infrastructure, including remote instruments and supercomputer clusters, (2) has ever-increasing throughput demands, and (3) has ever-increasing integration demands---for example, observations at one instrument could trigger a reconfiguration of another instrument. Today's DAQ transfers rely on UDP and (heavily tuned) TCP, but this is driven by convenience rather than suitability. The mismatch between Internet transport protocols and scientific workloads becomes more stark with the steady increase in link capacities, data generation, and integration across research infrastructure.This position paper argues the importance of developing specialized transport protocols for DAQ workloads. It proposes a new transport feature for this kind of elephant flow: multi-modality involves the network actively configuring the transport protocol to change how DAQ flows are processed across different underlying networks that connect scientific research infrastructure. Multi-modality is a layering violation that is proposed as a pragmatic technique for DAQ transport protocol design. It takes advantage of programmable network hardware that is increasingly being deployed in scientific research infrastructure. The paper presents an initial evaluation through a pilot study that includes a Tofino2 switch and Alveo FPGA cards, and using data from a particle detector.

97 MATHEMATICS AND COMPUTING↗

Flexible AI Models for Grid Resilience

The rapid growth in size and complexity of artificial intelligence (AI) and machine learning (ML) models has led to increased energy demands, posing a threat to the reliability of the existing power grid. This project addresses the challenge of highly intermittent and energy-intensive inference workloads by (1) developing fidelity-adaptive neural networks capable of dynamic response to grid conditions and (2) integrating these networks with power flow simulations to assess their impact on power grid reliability. We will explore both top-down and bottom-up approaches to create hierarchies of submodels that provide a controlled trade-off between power draw and prediction accuracy. The top-down method utilizes NN pruning to reduce a flagship model into progressively smaller, energy-efficient variants. The bottom-up approach employs geometrically principled weight setting strategies to construct depth-efficient models from the ground up. A real-time hardware-in-the-loop (HIL) platform will be developed to simulate a scaled AC power grid, integrating live AI workload power draw and enabling dynamic model switching in response to grid feedback. This work will provide a novel framework for evaluating the impact of flexible AI/ML workloads on grid performance and establish new methodologies for energy-aware computing in data centers. The outcomes will demonstrate that adaptive AI/ML can play a critical role in improving grid stability while advancing NREL's leadership in energy-efficient computing research.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Accelerating k-NN Algorithm with Hybrid MPI and OpenSHMEM

Machine Learning algorithms are benefiting from the continuous improvement of programming models, including MPI, MapReduce and PGAS. k-Nearest Neighbors (k-NN) algorithm is a widely used machine learning algorithm, applied to supervised learning tasks such as classification. Several parallel implementations of k-NN have been proposed in the literature and practice. However, on high-performance computing systems with high-speed interconnects, it is important to further accelerate existing designs of the k-NN algorithm through taking advantage of scalable programming models. To improve the performance of k-NN on large-scale environment with InfiniBand network, this paper proposes several alternative hybrid MPI+OpenSHMEM designs and performs a systemic evaluation and analysis on typical workloads. The hybrid designs leverage the one-sided memory access to better overlap communication with computation than the existing pure MPI design, and propose better schemes for efficient buffer management. The implementation based on k-NN program from MaTEx with MVAPICH2-X (Unified MPI+PGAS Communication Runtime over InfiniBand) shows up to 9.0% time reduction for training KDD Cup 2010 workload over 512 cores, and 27.6% time reduction for small workload with balanced communication and computation. Experiments of running with varied number of cores show that our design can maintain good scalability.

Lin, Jian↗

Synapse: Synthetic application profiler and emulator

Abstract Motivated by the need to emulate workload execution characteristics on high- performance and distributed heterogeneous resources, we introduce Synapse. Synapse is used as a proxy application (or "representative application") for real workloads, with the advantage that it can be tuned in different ways and dimensions, and also at levels of granularity that are not possible with real applications. Synapse has a platform-independent application profiler, and has the ability to emulate profiled workloads on a variety of resources. Experiments show that the automated profiling performed using Synapse captures an application's characteristics with high fidelity. The emulation of an application using Synapse can reproduce the application's execution behavior in the original run- time environment, and can also reproduce those behaviors on different run-time environments.

97 MATHEMATICS AND COMPUTING↗

Application configuration selection for energy-efficient execution on multicore systems

Balanced performance and energy consumption are incorporated in the design of modern computer systems. Several runtime factors, such as concurrency levels, thread mapping strategies, and dynamic voltage and frequency scaling (DVFS) should be considered in order to achieve optimal energy efficiency fora workload. Selecting appropriate run-time factors, however, is one of the most challenging tasks because the run-time factors are architecture-specific and workload-specific. And while most existing works concentrate on either static analysis of the workload or run-time prediction results, we present a hybrid two-step method that utilizes concurrency levels and DVFS settings to achieve the energy efficiency configuration for a worldoad. The experimental results based on a Xeon E5620 server with NPB and PARSEC benchmark suites show that the model is able to predict the energy efficient configuration accurately. On average, an additional 10% EDP (Energy Delay Product) saving is obtained by using run-time DVFS for the entire system. An off-line optimal solution is used to compare with the proposed scheme. Finally, the experimental results show that the average extra EDP saved by the optimal solution is within 5% on selective parallel benchmarks.

97 MATHEMATICS AND COMPUTING↗

Exploring performance and energy tradeoffs for irregular applications: A case study on the Tilera many-core architecture

High performance, parallel applications with irregular data accesses are becoming a critical workload class for modern systems. In particular, the execution of such workloads on emerging many-core systems is expected to be a significant component of applications in data mining, machine learning, scientific computing and graph analytics. However, power and energy constraints limit the capabilities of individual cores, memory hierarchy and on-chip interconnect of such systems, thus leading to architectural and software trade-os that must be understood in the context of the intended application’s behavior. Irregular applications are notoriously hard to optimize given their data-dependent access patterns, lack of structured locality and complex data structures and code patterns. We have ported two irregular applications, graph community detection using the Louvain method (Grappolo) and high-performance conjugate gradient (HPCCG), to the Tilera many-core system and have conducted a detailed study of platform-independent and platform-specific optimizations that improve their performance as well as reduce their overall energy consumption. To conduct this study, we employ an auto-tuning based approach that explores the optimization design space along three dimensions - memory layout schemes, GCC compiler flag choices and OpenMP loop scheduling options. We leverage MIT’s OpenTuner auto-tuning framework to explore and recommend energy optimal choices for different combinations of parameters. We then conduct an in-depth architectural characterization to understand the memory behavior of the selected workloads. Finally, we perform a correlation study to demonstrate the interplay between the hardware behavior and application characteristics. Using auto-tuning, we demonstrate whole-node energy savings and performance improvements of up to 49:6% and 60% relative to a baseline instantiation, and up to 31% and 45:4% relative to manually optimized variants.

auto tuning↗

An empirical study of I/O separation for burst buffers in HPC systems

To meet the exascale I/O requirements for the High-Performance Computing (HPC), a new I/O subsystem, Burst Buffer, based on solid state drives (SSD), has been developed. However, the diverse HPC workloads and the bursty I/O pattern cause severe data fragmentation that requires costly garbage collection (GC) and increases the number of bytes written to the SSD. To address this data fragmentation challenge, a new multi-stream feature has been developed for SSDs. In this work, we develop an I/O Separation scheme called BIOS to leverage this multi-stream feature to group the I/O streams based on the user IDs. We propose a stream-aware scheduling policy based on burst buffer pools in the workload manager, and integrate the BIOS with the workload manager to optimize the I/O separation scheme in burst buffer. We evaluate the proposed framework with a burst buffer I/O traces from Cori Supercomputer including a diverse set of applications. Experimental results show that the BIOS could improve the performance by 1.44x on average and reduce the Write Amplification Factor (WAF) by up to 1.20x. Finally, these demonstrate the potential benefits of the I/O separation scheme for solid state storage systems.

97 MATHEMATICS AND COMPUTING↗

Evaluating MPI resource usage summary statistics

The Message Passing Interface (MPI) remains the dominant programming model for scientific applications running on today’s high-performance computing (HPC) systems. This dominance stems from MPI’s powerful semantics for inter-process communication that has enabled scientists to write applications for simulating important physical phenomena. MPI does not, however, specify how messages and synchronization should be carried out. Those details are typically dependent on low-level architecture details and the message characteristics of the application. Therefore, analyzing an application’s MPI resource usage is critical to tuning MPI’s performance on a particular platform. The result of this analysis is typically a discussion of the mean message sizes, queue search lengths and message arrival times for a workload or set of workloads. While a discussion of the arithmetic mean in MPI resource usage might be the most intuitive summary statistic, it is not always the most accurate in terms of representing the underlying data. In this paper, we analyze MPI resource usage for a number of key MPI workloads using an existing MPI trace collector and discrete-event simulator. Our analysis demonstrates that the average, while easy and efficient to calculate, is a useful metric for characterizing latency and bandwidth measurements, but may not be a good representation of application message sizes, match list search depths, or MPI inter-operation times. Additionally, we show that the median and mode are superior choices in many cases. We also observe that the arithmetic mean is not the best representation of central tendency for data that are drawn from distributions that are multi-modal or have heavy tails. Furthermore, the results and analysis of our work provide valuable guidance on how we, as a community, should discuss and analyze MPI resource usage data for scientific applications.

97 MATHEMATICS AND COMPUTING↗

Optimization of distributed compute resources utilization in the CMS Global Pool

The CMS Submission Infrastructure is the primary system for managing computing resources for CMS workflows, including data processing, simulation, and analysis. It integrates geographically distributed resources from Grid, HPC, and cloud providers into federated pools managed by HTCondor and Glidein- WMS, for a total of around 500k CPU cores. This system dynamically manages workloads based on priorities defined by the collaboration. Additionally, CMS scheduling strategies must be flexible to handle multiple concurrent workloads while considering changing processing demands and resource availability from various providers.Efficient utilization of vast amounts of distributed compute resources is a key element for the success of the scientific programs of the LHC experiments. Optimizing the system is essential to maximize resource efficiency and fully utilize the distributed computing power. The CMS Submission Infrastructure team thus systematically investigates sources of inefficiency in workload scheduling to reduce their impact. In addition, a strategy of pilot overloading has been introduced to compensate for other inefficiency sources, thereby optimizing resource utilization and enhancing computational throughput.

Mascheroni, Marco [UC, San Diego (main)]↗

Data center growth in the United States: Decoupling the demand for services from electricity use

Data centers are energy intensive buildings that have grown in size and number to meet the increasing demands of a digital economy. This paper presents a bottom-up model to estimate data center electricity demand in the United States over a 20 year period and examines observed and projected electricity use trends in the context of changing data center operations. Results indicate a rapidly increasing electricity demand at the turn of the century that has significantly subsided to a nearly steady annual electricity use of about 70 billion kWh in recent years. While data center workloads continue to grow exponentially, comparable increases in electricity demand have been avoided through the adoption of key energy efficiency measures and a shift towards large cloud-based service providers. Alternative projections from the model illustrate the wide range in potential electricity that could be consumed to support data centers, with the US data center workload demand estimated for 2020 requiring a total electricity use that varies by about 135 billion kWh, depending on the adoption rate of efficiency measures during this decade. While recent improvements in data center energy efficiency have been a success, the growth of data center electricity use beyond 2020 is uncertain, as modeled trends indicate that the efficiency measures of the past may not be enough for the data center workloads of the future. The results show that successful stabilization of data center electricity will require new innovations in data center efficiency to further decouple electricity demand from the ever-growing demand for data center services.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

Adaptive Runtime Features For Distributed Graph Algorithms

Performance of distributed graph algorithms can benefit greatly by forming rapport between algorithmic abstraction and the underlying runtime system that is responsible for scheduling work and exchanging messages. However, due to their dynamic and irregular nature of computation, distributed graph algorithms written in different programming models impose varying degree of workload pressure on the runtime. To cope with such vastly different workload characteristics, a runtime has to make several trade-offs. One such trade-off arises, for example, when the runtime scheduler has to choose among alternatives such as whether to execute algorithmic work, or progress the network by probing network buffers, or throttle sending messages (termed flow control). This tradeoff decides between optimizing the throughput of a runtime scheduler by increasing the rate of execution of algorithmic work, and reducing the latency of the network messages. Another trade-off exists when a decision has to be made about when to send aggregated messages in buffers (message coalescing). This decision chooses between trading off latency for network bandwidth and vice versa. At any instant, such trade-offs emphasize either on improving the quantity of work being executed (by maximizing the scheduler throughput) or on improving the quality of work (by prioritizing better work). However, encoding static policies for different runtime features (such as flow control, coalescing) can prevent graph algorithms from achieving their full potential, thus can undermine the actual performance of a distributed graph algorithm . In this paper, we investigate runtime support for distributed graph algorithms in the context of two paradigms: variants of wellknown Bulk-Synchronous Parallel model and asynchronous programming model. We explore generic runtime features such as message coalescing (aggregation) and flow control and show that execution policies of these features need to be adjusted over time to make a positive impact on the execution time of a distributed graph algorithm. Since synchronous and asynchronous graph algorithms have different workload characteristics, not all of such runtime features may be good candidates for adaptation. Each of these algorithmic paradigms may require different set of features to be adapted over time. We demonstrate which set of feature(s) can be useful in each case to achieve the right balance of work in the runtime layer. Existing implementation of different graph algorithms can benefit from adapting dynamic policies in the underlying runtime.

Firoz, Jesun S.↗

Coding the Computing Continuum: Fluid Function Execution in Heterogeneous Computing Environments

Advances in network technologies have greatly decreased barriers to accessing physically distributed computers. This newfound accessibility coincides with increasing hardware specialization, creating exciting new opportunities to dispatch workloads to the best resource for a specific purpose, rather than those that are closest or most easily accessible. We present Delta, a service designed to intelligently schedule function-based workloads across a distributed set of heterogeneous computing resources. Delta implements an extensible architecture in which different predictors and scheduling algorithms can be integrated to provide dynamically evolving estimates of function execution times on different resources-estimates that can be used to determine the most appropriate location for execution. We describe predictors for function runtime, data transfer time, and cold-start resource provisioning and configuration delay; dynamic learning methods that update predictor models over time; and scheduling strategies that take into account both function and endpoint information. We show that these methods can halve workload makespan when compared with a strategy that selects the fastest resource, and decrease makespan by a factor of five when compared to a round robin strategy, when deployed on a heterogeneous testbed with resources ranging from a Raspberry Pi to a GPU node in an academic cloud.

Computing continuum↗

Analyzing a Five-Year Failure Record of a Leadership-Class Supercomputer

Extreme-scale computing systems are required to solve some of the grand challenges in science and technology. From astrophysics to molecular biology, supercomputers are an essential tool to accelerate scientific discovery. However, large computing systems are prone to failures due to their complexity. It is crucial to develop an understanding of how these systems fail to design reliable supercomputing platforms for the future. This paper examines a five-year failure and workload record of a leadership-class supercomputer. To the best of our knowledge, five years represents the vast majority of the lifespan of a supercomputer. This is the first time such analysis is performed on a top 10 modern supercomputer. We performed a failure categorization and found out that: i) most errors are GPU-related, with roughly 37% of them being double-bit errors on the cards; ii) failures are not evenly spread across the physical machine, with room temperature presumably playing a major role; and iii) software errors of the system bring down several nodes concurrently. Our failure rate analysis unveils that: i) the system consistently degrades, being at least twice as reliable at the beginning, compared to the end of the period; ii) Weibull distribution closely fits the mean-time-between-failure data; and iii) hardware and software errors show a markedly different pattern. Finally, we correlated failure and workload records to reveal that: i) failure and workload records are weakly correlated, except for certain types of failures when segmented by the hours of the day; ii) several categories of failures make jobs crash within the first minutes of execution; and iii) a significant fraction of failed jobs exhaust the requested time with a disregard of when the failure occurred during execution.

Rojas, Elvis↗

An Evaluation of the Effect of Network Cost Optimization for Leadership Class Supercomputers

Dragonfly-based networks are an extensively deployed network topology in large-scale high-performance computing due to their cost-effectiveness and efficiency. The US will soon have three Exascale supercomputers for leadership class workloads deployed using dragonfly networks. Compared to indirect networks of similar scale, the dragonfly network has considerably reduced cable lengths, cable counts, and switch counts, resulting in significant network cost savings for a given system size, however, these cost reductions result in reduced global minimal paths and more challenging routing. Additionally, large scale dragonfly networks often require a taper at the global link level, resulting in less bisection bandwidth than is achievable in other traditional non-blocking topologies of equivalent scale. While dragonfly networks have been extensively studied, they have yet to be fully evaluated in an extreme scale (i.e., exascale) system that targets capability workloads. In this paper, we present the results of the first large scale evaluation of a dragonfly network on an exascale system (Frontier) and compare its behavior to a similar scale fat-tree network on a previous generation TOP500 system (Summit). This evaluation aims to determine the effect of network cost optimizations by measuring a tapered topology’s impact on capability workloads. Our evaluation is based on a collection of synthetic microbenchmarks, mini-apps, and full scale applications. It compares the scaling efficiencies of each benchmark between the dragonfly-based Frontier and the fat-tree-based Summit systems. Our results show that a dragonfly network is $\sim \mathbf{3 0 \%}$ more cost efficient than a fat-tree topology, which amortizes to $\sim 3 \%$ of an exascale system cost. Furthermore, while tapered dragonfly networks impose significant tradeoffs, the impacts are not as broad as initially thought and are mostly seen in applications with global communication patterns, particularly all-to-all (e.g., FFT-based algorithms), but also local communication patterns (e.g., nearest-neighbor algorithms) that are sensitive to network performance variability.

Khan, Awais↗

FPDeep: Scalable Acceleration of CNN Training on Deeply-Pipelined FPGA Clusters

In this paper, we propose a framework called FPDeep, which uses a hybrid of model and layer paral- lelism to configure distributed reconfigurable clusters to train DNNs. This approach has numerous benefits. First, the design does not suffer from batch size growth. Second, novel workload and weight partitioning leads to balanced loads of both among nodes. And third, the entire system is fine-grained pipeline. This leads to high parallelism and utilization and also minimizes the time features need to be cached while waiting for back-propagation.

Wang, Tianqi↗