Workload Characteristics Research Articles

Optimizing IT FinOps and Sustainability through Unsupervised Workload Characterization

The widespread adoption of public and hybrid clouds, along with elastic resources and various automation tools for dynamic deployment, has accelerated the rapid provisioning of compute resources as needed. Despite these advancements, numerous resources persist unnecessarily due to factors such as poor digital hygiene, risk aversion, or the absence of effective tools, resulting in substantial costs and energy consumption. Existing threshold-based techniques prove inadequate in effectively addressing this challenge. To address this issue, we propose an unsupervised machine learning framework to automatically identify resources that can be de-provisioned completely or summoned on a schedule. Application of this approach to enterprise data has yielded promising initial results, facilitating the segregation of productive workloads with recurring demands from non-productive ones.

WattScope: Non-intrusive Application-level Power Disaggregation in Datacenters

WattScope is a system for non-intrusively estimating the power consumption of individual applications using external measurements of a server's aggregate power usage and without requiring direct access to the server's operating system or applications. Our key insight is that, based on an analysis of production traces, the power characteristics of datacenter workloads, e.g., low variability, low magnitude, and high periodicity, are highly amenable to disaggregation of a server's total power consumption into application-specific values. WattScope adapts and extends a machine learning-based technique for disaggregating building power and applies it to server- and rack-level power measurements that are already available in datacenters. We evaluate WattScope's accuracy on a production workload and show that it yields high accuracy, e.g., often <∼10% normalized mean absolute error, and is thus a potentially useful tool for datacenters in externally monitoring application-level power usage.

Data Placement Using a Classifier for SLC/QLC Hybrid SSDs

In hybrid SSDs (solid-state drives) consisting of SLC (single-level cell) and QLC (quad-level cell), efficiently using the limited SLC cache space is crucial. In this paper, we present a practical data placement scheme, which determines the placement location of incoming write requests using a lightweight machine-learning model. It leverages information about I/O workload characteristics and SSD status to identify cold data that does not need to be stored in the SLC cache with high accuracy. By strategically bypassing the SLC cache for cold data, our scheme significantly reduces unnecessary data movements between the SLC and QLC regions, improving the overall efficiency of the SSD. Through simulation-based studies using real-world workloads, we demonstrate that our scheme outperforms existing approaches by up to 44%.

POLAR: Adaptive and Non-invasive Join Order Selection via Plans of Least Resistance

Join ordering and query optimization are crucial for query performance but remain challenging due to unknown or changing characteristics of query intermediates, especially for complex queries with many joins. Over the past two decades, a spectrum of techniques for adaptive query processing (AQP)---including inter-/intra-operator adaptivity and tuple routing---have been proposed to address these challenges. However, commercial database systems in practice do not implement holistic AQP techniques because they increase the system complexity (e.g., intertwined planning and execution) and thus, complicate debugging and testing. Additionally, existing approaches may incur large overheads, leading to problematic performance regressions. In this paper, we introduce POLAR, a simple yet very effective technique for a self-regulating selection of alternative join orderings with bounded overhead. We enhance left-deep join pipelines with alternative join orders, perform regret-bounded tuple routing to find and validate "plans of least resistance", and then process the majority of tuple batches through these plans. We study different join order selection techniques, different routing strategies, and a variety of workload characteristics. Our experiments with a POLAR prototype in DuckDB show runtime improvements of up to 9x and less than 7% overhead for all benchmark queries, while outperforming state-of-the-art AQP systems by up to 15x.

Efficient Cross-platform Multiplexing of Hardware Performance Counters via Adaptive Grouping

Collecting sufficient microarchitecture performance data is essential for performance evaluation and workload characterization. There are many events to be monitored in a modern processor while only a few hardware performance monitoring counters (PMCs) can be used, so multiplexing is commonly adopted. However, inefficiency commonly exists in state-of-the-art profiling tools when grouping events for multiplexing PMCs. It has the risk of inaccurate measurement and misleading analysis. Commercial tools can leverage PMCs, but they are closed source and only support their specified platforms. To this end, we propose an approach for efficient cross-platform microarchitecture performance measurement via adaptive grouping, aiming to improve the metrics’ sampling ratios. The approach generates event groups based on the number of available PMCs detected on arbitrary machines while avoiding the scheduling pitfall of Linux perf_event subsystem. We evaluate our approach with SPEC CPU 2017 on four mainstream x86-64 and AArch64 processors and conduct comparative analyses of efficiency with two other state-of-the-art tools, LIKWID and ARM Top-down Tool. The experimental results indicate that our approach gains around 50% improvement in the average sampling ratio of metrics without compromising the correctness and reliability.

Optimizing Task Scheduling and Resource Allocation in Computing Environments using Metaheuristic Methods

Optimizing system performance in dynamic and heterogeneous environments and the efficient management of computational tasks are crucial. This paper therefore looks at task scheduling and resource allocation algorithms in some depth. The work evaluates five algorithms: Genetic Algorithms (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Firefly Algorithm (FA) and Simulated Annealing (SA) across various workloads achieved by varying the task-to-node ratio. The paper identifies Finish Time and Deadline as two key performance metrics for gauging the efficacy of an algorithm, and a comprehensive investigation of the behaviors of these algorithms across different workloads was carried out. Results from the experiments reveal unique patterns in algorithmic behaviors by workload. In the 15-task and 5-node scenario, the GA and PSO algorithms outclass all others, completing 100 percent of tasks before deadlines, Task 5 was a bane to the ACO algorithm. The study proposes a more extensive system that promotes an adaptive algorithmic approach based on workload characteristics. Numerically, the GA and PSO algorithms triumphed completing 100 percent of tasks before their deadlines in the face of 10 tasks and 5 nodes, while the ACO algorithm stumbled on certain tasks. As it is stated in the study, The above-mentioned system offers an integrated approach to ill-structured problem of task scheduling and resource allocation. It offers an intelligent and aggressive scheduling scheme that runs asynchronously when a higher number of tasks is submitted for the completion in addition to those dynamically aborts whenever system load and utilization cascade excessively. The proposed design seems like full-fledged solution over project scheduling or resource allocation issues. It highlights a detailed method of the choice of algorithms based on semantic features, aiming at flexibility. Effects of producing quantifiable statistical results from the experiments on performance empirically demonstrate each algorithm performed under various settings.

Characterization and Machine Learning Classification of AI and PC Workloads

Characterization and Machine Learning Classification of AI and PC Workloads

Investigation of time-motion characteristics of work load and physiological load in different regimes of small-sided games in soccer

Investigation of time-motion characteristics of work load and physiological load in different regimes of small-sided games in soccer

A Workload Characterization Methodology Using Supervised and Unsupervised Deep Learning

A Workload Characterization Methodology Using Supervised and Unsupervised Deep Learning

Efficient Workload Allocation and Scheduling Strategies for AI-Intensive Tasks in Cloud Infrastructures.

- The rapid proliferation of Artificial Intelligence (AI) applications has underscored the need for advanced cloud infrastructures capable of efficiently managing AI-intensive workloads. This paper delves into the intricacies of workload allocation and scheduling in the context of cloud environments, specifically focusing on the challenges posed by AI-intensive tasks. Our research endeavors to scrutinize existing strategies, discern their limitations, and proffer innovative approaches tailored to optimize the allocation and scheduling of AI workloads within cloud infrastructures. In elucidating the challenges, we pinpoint resource heterogeneity, dynamic workload characteristics, and scalability as the crux of the issues confronting AI-intensive workload management. The diverse computational demands of AI workloads make it challenging to allocate resources optimally, while the dynamic nature of these tasks necessitates adaptive strategies to accommodate varying computational requirements over time. [1] Additionally, as AI models and datasets burgeon in complexity and size, ensuring scalability becomes paramount for sustaining performance in cloud environments. Our literature review encompasses an examination of both traditional and state-of-the-art workload allocation strategies, shedding light on their respective strengths and shortcomings. We also delve into scheduling techniques employed for managing AI-intensive tasks, providing a comprehensive overview of the existing landscape. To address these challenges, we propose a novel framework centered around dynamic resource provisioning, machine learning-based scheduling, and efficient task migration strategies. The framework aims to adaptively allocate resources based on the evolving nature of AI workloads, leveraging machine learning algorithms to predict workload characteristics and employing efficient task migration to handle workload fluctuations. The paper concludes with an experimental evaluation of the proposed strategies, conducted in a simulated environment using diverse datasets. Key performance metrics, such as throughput, latency, and resource utilization, are employed to assess the effectiveness of our strategies compared to existing approaches. By offering insights into the efficient management of AI-intensive workloads in cloud infrastructures, this research contributes to the ongoing efforts to enhance the scalability and performance of cloud environments in the face of burgeoning AI applications.

Analysis of Academic Web Server Traffic and Workload Characterization for Performance Evaluation

Analysis of Academic Web Server Traffic and Workload Characterization for Performance Evaluation

Effects of cognitive load and different exercise intensities on perceived effort in sedentary university students: a follow up of the Cubo Fitness Test validation.

Work and intellectually fatiguing environments can significantly influence the health of individuals, which is strictly bound to motor efficiency. In particular, desk workers and university students may have a sedentary lifestyle and a condition of mental fatigue caused by daily routine, which could impair motor efficiency. The assessment is a starting point for enhancing awareness of the individual's psychophysical condition through the perception of one's body motor efficiency, motivating to move towards improvement. This way, a submaximal test based on perceived exertion was developed (Cubo Fitness Test, CFT) and validated in previous studies. Hence, two further studies were employed to enhance the consistency and accuracy of this instrument in different conditions. The first study investigated the internal responsiveness of CFT, evaluating if mental fatigue could affect motor efficiency. The second study investigated which perceived intensity (weak, moderate, strong, or absolute maximum) could be more reliable for applying the CFT (as previous research focused the investigation only on moderate intensity). In the first investigation, participants assessed two stimuli (mental fatigue induced with a Stroop color-word task and a neutral condition based on the vision of a documentary) lasting 60 min each. The quality of psychophysical recovery (total quality recovery) and the mood state (Italian Mood State questionnaire) were evaluated before the stimuli. After the fatiguing or the neutral task, the mood state was newly assessed, together with the evaluation of the workload's characteristics (Nasa TLX) and the CFT motor efficiency. In the second investigation, participants had to perform CFT twice for each at different intensities of Borg's Scale of perceived exertion. Researchers successfully requested to fill out the NASA TLX questionnaire regarding the perceived workload characteristics of CFT, and the reliability of each intensity was assessed. Results seem to enhance the consistency and the accuracy of the instrument. Indeed, findings evidenced that CFT is not influenced by mental fatigue conditions typical of the intellectual work of desk workers and university students for which this test was specifically conceived. Moreover, moderate and strong perceived intensity are the most adequate conditions to assess motor efficiency in these populations.

Improving graphics processing unit performance based on neural network direct memory access controller

In this paper proposes the design and implementation of the back propagation algorithm based neural network DMA (Direct Memory Access) Controller for use of multimedia applications. The proposed DMA controller work with the back propagation-training algorithm. The advantages of the back propagation algorithm it will be work on the gradient loss w.r.t the network weights. So this back propagation algorithm is used as training algorithm for the DMA controller. The proposed method is test with the different workload characteristics like heavy workload, medium workload and normal workload. The performance parameters are considered here is like accuracy, precision, recall and F1 score etc. The proposed method is compared with existing methods like CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), LSTM (Long Sort term Memory) and GRU (Gated Recurrent Unit) etc. Finally, the proposed design will give the better performance than existing methods.

Speed behaviour and mental workload of small-spacing expressway interchanges based on field driving test

Many small-spacing interchanges (SSI) appear with the improvement of the expressway network. To investigate the speed and mental workload characteristics in the SSI and acquire the mechanism of the influence of speed on the drivers’ workload, 37 participants were recruited to perform a field driving test. Each driver performed four driving conditions (i.e. ramp-mainline, mainline-ramp, mainline driving, and auxiliary lane driving). The speed and drivers’ electrocardiogram (ECG) data were collected using SpeedBox speed acquisition equipment and PhysioLAB physiological instrument. The heart rate increase (HRI) index was used to analyse the drivers’ mental workload regularity. The relationship model between speed and HRI was developed to examine the impact of speed on HRI. The results show that the speed variation in the SSI displayed two patterns: ‘decrease - increase and continuous decrease.’ The drivers’ HRI variation presented four patterns: ‘convex curve, continuously increasing, continuously decreasing and concave curve’. SSI's influenced area length is given based on the speed and HRI variation regularity. HRI is significantly higher when driving in the ramp-mainline condition in the SSI than when driving in other conditions, indicating that drivers are more nervous when merging with the mainline traffic. HRI increases significantly in the first 50% of the weaving area in four driving conditions, indicating that vehicle weaving greatly influences the drivers’ mental workload. A positive correlation exists between vehicle speed and drivers’ HRI without interference from other vehicles and road alignment.

Analyzing Data Reference Characteristics of Deep Learning Workloads for Improving Buffer Cache Performance

Due to the recent growing data size of deep learning workloads, loading data from storage is increasingly becoming a performance bottleneck for neural network systems. In this article, we analyze the data reference characteristics of neural network workloads and observe that they are significantly different from conventional desktop workloads. In particular, during the training phase of deep learning, data blocks are referenced in a fully random manner, which significantly degrades the performance of a buffer cache. To handle this situation, this article suggests a new data shuffling scheme that aims to accelerate data loading in deep neural networks. Unlike the default shuffling method used in PyTorch that randomly shuffles full dataset in every epoch, the proposed scheme defines a shuffling unit called bundle, and enhances the locality of data references to improve buffer cache performances. Specifically, the proposed scheme performs data shuffling by the unit of a bundle, and the bundles used in each epoch are arranged alternately, thereby improving the locality of references at the viewpoint of the buffer cache. Based on simulation and measurement studies, we show that the hit rate of the buffer cache is improved by 37.2%, and the data loading time is also shortened by 11.4% without degrading the model’s training efficiency.

Development of a Neuroergonomic Assessment for the Evaluation of Mental Workload in an Industrial Human–Robot Interaction Assembly Task: A Comparative Case Study

The disruptive deployment of collaborative robots, named cobots, in Industry 5.0 has brought attention to the safety and ergonomic aspects of industrial human–robot interaction (HRI) tasks. In particular, the study of the operator’s mental workload in HRI activities has been the research object of a new branch of ergonomics, called neuroergonomics, to improve the operator’s wellbeing and the efficiency of the system. This study shows the development of a combinative assessment for the evaluation of mental workload in a comparative analysis of two assembly task scenarios, without and with robot interaction. The evaluation of mental workload is achieved through a combination of subjective (NASA TLX) and real-time objective measurements. This latter measurement is found using an innovative electroencephalogram (EEG) device and the characterization of the cognitive workload through the brainwave power ratio β/α, defined after the pre-processing phase of EEG data. Finally, observational analyses are considered regarding the task performance of the two scenarios. The statistical analyses show how significantly the mental workload diminution and a higher level of performance, as the number of components assembled correctly by the participants, are achieved in the scenario with the robot.

WattScope: Non-intrusive application-level power disaggregation in datacenters

Datacenter capacity is growing exponentially to satisfy the increasing demand for many emerging computationally-intensive applications, such as deep learning. This trend has led to concerns over datacenters’ increasing energy consumption and carbon footprint. The most basic prerequisite for optimizing a datacenter’s energy- and carbon-efficiency is accurately monitoring and attributing energy consumption to specific users and applications. Since datacenter servers tend to be multi-tenant, i.e., they host many applications, server- and rack-level power monitoring alone does not provide insight into the energy usage and carbon emissions of their resident applications. At the same time, current application-level energy monitoring and attribution techniques are intrusive: they require privileged access to servers and necessitate coordinated support in hardware and software, neither of which is always possible in cloud environments. To address the problem, we design WattScope, a system for non-intrusively estimating the power consumption of individual applications using external measurements of a server’s aggregate power usage and without requiring direct access to the server’s operating system or applications. Our key insight is that, based on an analysis of production traces, the power characteristics of datacenter workloads, e.g., low variability, low magnitude, and high periodicity, are highly amenable to disaggregation of a server’s total power consumption into application-specific values. WattScope adapts and extends a machine learning-based technique for disaggregating building power and applies it to server- and rack-level power meter measurements that are already available in data centers. We evaluate WattScope’s accuracy on a production workload and show that it yields high accuracy, e.g., often <∼10% normalized mean absolute error, and is thus a potentially useful tool for datacenters in externally monitoring application-level power usage.

EMS-i : An Efficient Memory System Design with Specialized Caching Mechanism for Recommendation Inference

Recommendation systems have been widely embedded into many Internet services. For example, Meta’s deep learning recommendation model (DLRM) shows high prefictive accuracy of click-through rate in processing large-scale embedding tables. The SparseLengthSum (SLS) kernel of the DLRM dominates the inference time of the DLRM due to intensive irregular memory accesses to the embedding vectors. Some prior works directly adopt near data processing (NDP) solutions to obtain higher memory bandwidth to accelerate SLS. However, their inferior memory hierarchy induces low performance-cost ratio and fails to fully exploit the data locality. Although some software-managed cache policies were proposed to improve the cache hit rate, the incurred cache miss penalty is unacceptable considering the high overheads of executing the corresponding programs and the communication between the host and the accelerator. To address the issues aforementioned, we propose EMS-i , an efficient memory system design that integrates Solide State Drive (SSD) into the memory hierarchy using Compute Express Link (CXL) for recommendation system inference. We specialize the caching mechanism according to the characteristics of various DLRM workloads and propose a novel prefetching mechanism to further improve the performance. In addition, we delicately design the inference kernel and develop a customized mapping scheme for SLS operation, considering the multi-level parallelism in SLS and the data locality within a batch of queries. Compared to the state-of-the-art NDP solutions, EMS-i achieves up to 10.9× speedup over RecSSD and the performance comparable to RecNMP with 72% energy savings. EMS-i also saves up to 8.7× and 6.6 × memory cost w.r.t. RecSSD and RecNMP, respectively.

Dynamic Power Management in Large Manycore Systems: A Learning-to-Search Framework

The complexity of manycore System-on-chips (SoCs) is growing faster than our ability to manage them to reduce the overall energy consumption. Further, as SoC design moves toward three-dimensional (3D) architectures, the core's power density increases leading to unacceptable high peak chip temperatures. In this article, we consider the optimization problem of dynamic power management (DPM) in manycore SoCs for an allowable performance penalty (say, 5%) and admissible peak chip temperature. We employ a machine learning– (ML) based DPM policy, which selects the voltage/frequency levels for different cluster of cores as a function of the application workload features such as core computation and inter-core traffic, and so on. We propose a novel learning-to-search (L2S) framework to automatically identify an optimized sequence of DPM decisions from a large combinatorial space for joint energy-thermal optimization for one or more given applications. The optimized DPM decisions are given to a supervised learning algorithm to train a DPM policy, which mimics the corresponding decision-making behavior. Our experiments on two different manycore architectures designed using wireless interconnect and monolithic 3D demonstrate that principles behind the L2S framework are applicable for more than one configuration. Moreover, L2S-based DPM policies achieve up to 30% energy-delay product savings and reduce the peak chip temperature by up to 17 °C compared to the state-of-the-art ML methods for an allowable performance overhead of only 5%.

DecLog: Decentralized Logging in Non-Volatile Memory for Time Series Database Systems

Growing demands for the efficient processing of extreme-scale time series workloads call for more capable time series database management systems (TSDBMS). Specifically, to maintain consistency and durability of transaction processing, systems employ write-ahead logging (WAL) whereby transactions are committed only after the related log entries are flushed to disk. However, when faced with massive I/O, this becomes a throughput bottleneck. Recent advances in byte-addressable Non-Volatile Memory (NVM) provide opportunities to improve logging performance by persisting logs to NVM instead. Existing studies typically track complex transaction dependencies and use barrier instructions of NVM to ensure log ordering. In contrast, few studies consider the heavy-tailed characteristics of time series workloads, where most transactions are independent of each other. We propose DecLog, a decentralized NVM-based logging system that enables concurrent logging of TSDBMS transactions. Specifically, we propose data-driven log sequence numbering and relaxed ordering strategies to track transaction dependencies and resolve serialization issues. We also propose a parallel logging method to persist logs to NVM after being compressed and aligned. An experimental study on the YCSB-TS benchmark offers insight into the performance properties of DecLog, showing that it improves throughput by up to 4.6× while offering lower recovery time in comparison to the open source TSDBMS Beringei.