Memory Utilization Research Articles

Towards Resource Efficiency: Practical Insights into Large-Scale Spark Workloads at ByteDance

At ByteDance, where we execute over a million Spark jobs and handle 500PB of shuffled data daily, ensuring resource efficiency is paramount for cost savings. However, achieving optimization of resource efficiency in large-scale production environments poses significant challenges. Drawing from our practical experiences, we have identified three key issues critical to addressing resource efficiency in real-world production settings: 1 slow I/Os leading to excessive CPU and memory idleness, 2 coarse-grained resource control causing wastage, and 3 sub-optimal job configurations resulting in low utilization. To tackle these issues, we propose a resource efficiency governance framework for Spark workloads. Specifically, 1 we devise the multi-mechanism shuffle services, including Enhanced External Shuffle Service (ESS) and Cloud Shuffle Service (CSS), where CSS employs a push-based approach to enhance I/O efficiency through sequential reading. 2 We modify the Spark configuration parameter protocol, allowing for fine-grained resource control by introducing several new parameters such as milliCores and memoryBurst, as well as supporting operators with additional spill modes. 3 We design a two-stage configuration autotuning method, comprising rule-based and algorithm-based tuning, providing more reliable Spark configuration optimizations. By deploying these techniques on millions of Spark jobs in production over the last two years, we have achieved over 22% CPU utilization increase, 5% memory utilization increase, and 10% shuffle block time ratio decrease, effectively saving millions of CPU cores and petabytes of memory daily.

DLRover-RM: Resource Optimization for Deep Recommendation Models Training in the Cloud

Deep learning recommendation models (DLRM) rely on large embedding tables to manage categorical sparse features. Expanding such embedding tables can significantly enhance model performance, but at the cost of increased GPU/CPU/memory usage. Meanwhile, tech companies have built extensive cloud-based services to accelerate training DLRM models at scale. In this paper, we conduct a deep investigation of the DLRM training platforms at AntGroup and reveal two critical challenges: low resource utilization due to suboptimal configurations by users and the tendency to encounter abnormalities due to an unstable cloud environment. To overcome them, we introduce DLRover, an elastic training framework for DLRMs designed to increase resource utilization and handle the instability of a cloud environment. DLRover develops a resource-performance model by considering the unique characteristics of DLRMs and a three-stage heuristic strategy to automatically allocate and dynamically adjust resources for DLRM training jobs for higher resource utilization. Further, DLRover develops multiple mechanisms to ensure efficient and reliable execution of DLRM training jobs. Our extensive evaluation shows that DLRover reduces job completion times by 31%, increases the job completion rate by 6%, enhances CPU usage by 15%, and improves memory utilization by 20%, compared to state-of-the-art resource scheduling frameworks. DLRover has been widely deployed at AntGroup and processes thousands of DLRM training jobs on a daily basis. DLRover is open-sourced and has been adopted by 10+ companies.

A Novel Approach to Construct Finite Automata Using Grid and Product Automata

This research aims to utilize an organized grid-based strategy to make the development of complex Finite Automata easier. Product Automata are used to merge many automata into a single automaton to integrate various computing processes. Combining these methodologies provides novel methods of improving the scalability and efficiency of automata building, broadening the field of research for automata theory and its applications. The scalability of this developed automated system will benefit sectors such as automotive production and logistics. The results indicate considerable improvements in construction time, memory usage, scalability, and resilience compared to older approaches. The performance of the developed method, as measured by construction time, memory utilization, scalability, robustness, application range, and complexity, is 25% to 50% higher than that of traditional methodologies in the literature.

A novel memory-based artificial gorilla troops optimizer for installing biomass distributed generators in unbalanced radial networks

Optimizing unbalanced distribution networks through the strategic integration of distributed generators (DGs) has long been recognized as a significant challenge. Selecting the optimal sizes and locations for these generators is crucial for minimizing network power loss and enhancing voltage profiles. The previously published methods have been plagued by issues such as slow convergence rates, entrapment in local optima, complexity, and extensive computational requirements. Addressing these limitations, this paper introduces an efficient methodology: the Memory-based Artificial Gorilla Troops Optimizer (MGTO). This approach leverages memory-based mechanisms to enhance exploration and decision-making, facilitating the seamless integration of various biomass DGs (BDGs) into unbalanced IEEE 37-bus radial networks. The immigration of gorillas during the exploration phase is enriched through the utilization of stored memories of candidate trajectories within the search space, enabling the silverback to make informed decisions. Furthermore, a multi-objective variant of MGTO is developed in collaboration with Fuzzy Decision-Making (FDM), allowing for the simultaneous optimization of multiple targets. To demonstrate the MGTO effectiveness, it is rigorously compared against a comprehensive set of established optimization algorithms, including the Honey Badger Algorithm (HBA), Runge Kutta Optimizer (RUN), and others. The results proved the dominance of the proposed MGTO by getting minimum power loss and voltage fluctuation of 0.364 % and 15.4 %, respectively, while in the multi-objective problem, the best results are 0.513 % loss and 17.9% voltage fluctuation. The results proved the consistency of the proposed MGTO in installing different BDGs into an unbalanced distribution network.

A High-Performance and Ultra-Low-Power Accelerator Design for Advanced Deep Learning Algorithms on an FPGA

This article addresses the growing need in resource-constrained edge computing scenarios for energy-efficient convolutional neural network (CNN) accelerators on mobile Field-Programmable Gate Array (FPGA) systems. In particular, we concentrate on register transfer level (RTL) design flow optimization to improve programming speed and power efficiency. We present a re-configurable accelerator design optimized for CNN-based object-detection applications, especially suitable for mobile FPGA platforms like the Xilinx PYNQ-Z2. By not only optimizing the MAC module using Enhanced clock gating (ECG), the accelerator can also use low-power techniques such as Local explicit clock gating (LECG) and Local explicit clock enable (LECE) in memory modules to efficiently minimize data access and memory utilization. The evaluation using ResNet-20 trained on the CIFAR-10 dataset demonstrated significant improvements in power efficiency consumption (up to 22%) and performance. The findings highlight the importance of using different optimization techniques across multiple hardware modules to achieve better results in real-world applications.

Gesture Recognition of Filipino Sign Language Using Convolutional and Long Short-Term Memory Deep Neural Networks

In response to the recent formalization of Filipino Sign Language (FSL) and the lack of comprehensive studies, this paper introduces a real-time FSL gesture recognition system. Unlike existing systems, which are often limited to static signs and asynchronous recognition, it offers dynamic gesture capturing and recognition of 10 common expressions and five transactional inquiries. To this end, the system sequentially employs cropping, contrast adjustment, grayscale conversion, resizing, and normalization of input image streams. These steps serve to extract the region of interest, reduce the computational load, ensure uniform input size, and maintain consistent pixel value distribution. Subsequently, a Convolutional Neural Network and Long-Short Term Memory (CNN-LSTM) model was employed to recognize nuances of real-time FSL gestures. The results demonstrate the superiority of the proposed technique over existing FSL recognition systems, achieving an impressive average accuracy, recall, and precision rate of 98%, marking an 11.3% improvement in accuracy. Furthermore, this article also explores lightweight conversion methods, including post-quantization and quantization-aware training, to facilitate the deployment of the model on resource-constrained platforms. The lightweight models show a significant reduction in model size and memory utilization with respect to the base model when executed in a Raspberry Pi minicomputer. Lastly, the lightweight model trained with the quantization-aware technique (99%) outperforms the post-quantization approach (97%), showing a notable 2% improvement in accuracy.

Foraging trail traffic rules: a new study method of trajectories of the harvester ants.

Harvester ants are one of the most extensively studied groups of ants, especially the group foraging ants, Messor barbarus (Linnaeus, 1767), which construct long-lasting trunk trails. Limited laboratory investigations have delved into head-on encounters along foraging trails involving workers moving in opposing directions, with fewer corresponding studies conducted in the natural environment. To address this gap, we devised an in-field experimental design to induce lane segregation on the foraging trunk trail of M. barbarus. Using an image-based tracking method, we analyzed the foraging behavior of this species to assess the costs associated with head-on encounters and to figure out the natural coexistence of outgoing and returning workers on a bidirectional route. Our results consistently reveal heightened straightness and speed in unidirectional test lanes, accompanied by an elevated foraging rate compared to bidirectional lanes. This suggests a potential impact of head-on collisions on foraging behavior, especially on foraging efficiency. Additionally, Kinematic analysis revealed distinct movement patterns between outbound and inbound flows, particularly low speed and sinuous trajectories of inbounding unladen workers. The study of encounter rates in two traffic systems hints at the plausible utilization of individual memory by workers within trails, underscoring the pivotal role of encounters in information exchange and load transfer.

Architecture and iot security systems based on fog computing

The subject of the study is is the security architecture of the Internet of Things (IoT) based on fog computing, which allows providing efficient and secure services for many IoT users. The goal is to investigate the security architecture for IoT systems based on fog computing. To achieve the goal, the following tasks were solved: the concept of fog computing is proposed, its architecture is considered and a comparative analysis of fog and cloud computing architectures is made; the principles of designing and implementing the architecture of a fog computing system are outlined; multi-level security measures based on fog computing are investigated; and the areas of use of fog computing-based Internet of Things networks are described. When performing the tasks, such research methods were used as: theoretical analysis of literature sources; analysis of the principles of designing and implementing the security architecture of the Internet of Things; analysis of security measures at different levels of the architecture. The following results were obtained: the architecture of fog computing is considered and compared with the cloud architecture; the principles of designing and implementing the architecture of fog computing systems are formulated; multi-level IoT security measures based on fog computing are proposed. Conclusions: research on IoT security systems based on fog computing has important theoretical implications. The fog computing architecture, in contrast to the cloud architecture, better meets the demand for high traffic and low latency of mobile applications, providing more advantages for systems that require real-time information processing. When designing and implementing the architecture of fog computing systems, the factors of memory capacity, latency, and utility should be taken into account to effectively integrate fog technologies with IoT. To ensure a high level of system security, multi-level security measures should be implemented using both software and hardware solutions.

Automatically optimized component model computation for power system simulation on GPU

This work provides an approach that automatically optimizes the component computations on graphics processing unit (GPU) devices from different vendors. The approach consists of a two-level optimization, where the first level considers the linear part of the computation for vectorization and applies mixed matrix formats to increase computational throughput further. Then, the second optimization level treats the combination of linear and non-linear parts as a black box and searches for the optimal configuration of parameters such as the degree of vectorization, the combination of matrix formats, and the group (of threads) sizes during parallel execution on GPU. Moreover, we also introduce constraints that reduce the optimization procedure’s execution time. Finally, we select three different types of components that could be representative to computational tasks in power system and perform our optimization approach on these kernels. The computational performance is compared with unoptimized baseline and sparse linear algebra library based implementations, result shows that our optimization leads to better performance and more efficient memory utilization.

Global–Local Deep Fusion: Semantic Integration with Enhanced Transformer in Dual-Branch Networks for Ultra-High Resolution Image Segmentation

The fusion of global contextual information with local cropped block details is crucial for segmenting ultra-high resolution images. In this study, A novel fusion mechanism termed global–local deep fusion (GL-Deep Fusion) is introduced, based on an enhanced transformer architecture that efficiently integrates global contextual information and local details. Specifically, we propose the global–local synthesis networks (GLSNet), a dual-branch network where one branch processes the entire original image, while the other branch handles cropped local patches as input. The feature fusion of different branches in GLSNet is achieved through GL-Deep Fusion, significantly enhancing the accuracy of ultra-high resolution image segmentation. Identifying tiny overlapping items is a task where the model excels, demonstrating its particular effectiveness. To optimize GPU memory utilization, a dual-branch architecture was meticulously designed. This architecture proficiently leverages the features it extracts and seamlessly integrates them into the enhanced transformer framework of GL-Deep Fusion. Benchmarks on the DeepGlobe and Vaihingen datasets demonstrate the efficiency and accuracy of the proposed model. It significantly reduces GPU memory usage by 24.1% on the DeepGlobe dataset, enhancing segmentation accuracy by 0.8% over the baseline model. On the Vaihingen dataset, our model delivers a Mean F1 score of 90.2% and achieves a mIoU of 90.9%, highlighting its exceptional memory efficiency and segmentation precision.

A fast and highly scalable frequent pattern mining algorithm

As the use of big data and its potential benefits become more widespread, public and private organizations around the world have realized the imperative of incorporating comprehensive and robust technologies into their business processes. In particular, companies are implementing more and more intelligent systems into their business processes, resulting in an exponential increase in the amount of data being collected and used in everyday life on a quarterly basis. Therefore, an algorithm that can efficiently explore the frequent patterns from big data will be able to productively analyze and utilize the generated data to better optimize resources and even develop new business models. Thus, in this urban data era, it is not only essential to make more efficient and accurate decisions to improve business profitability and customer satisfaction, but also a highly challenging core issue. In this study, a new approach, scalable FP mining, is proposed for the first time to achieve a certain level of performance and more efficient memory utilization despite large amounts of data. Experiment results for different data characteristics show that the best performance of the proposed method requires only 48.9% of the DP algorithm's execution time. In further experiments with increased exploration difficulty, the optimal execution time is only 37%. On specific associated databases, the proposed method maintains stable and excellent performance with 12.7% for different data characteristics. The experimental results with increasing exploration difficulty at each stage demonstrate the stability, robustness, and reliability of the proposed SFP, especially under high data complexity and exploration difficulty.

Efficiency and Security Evaluation of Lightweight Cryptographic Algorithms for Resource-Constrained IoT Devices.

The IoT has become an integral part of the technological ecosystem that we all depend on. The increase in the number of IoT devices has also brought with it security concerns. Lightweight cryptography (LWC) has evolved to be a promising solution to improve the privacy and confidentiality aspect of IoT devices. The challenge is to choose the right algorithm from a plethora of choices. This work aims to compare three different LWC algorithms: AES-128, SPECK, and ASCON. The comparison is made by measuring various criteria such as execution time, memory utilization, latency, throughput, and security robustness of the algorithms in IoT boards with constrained computational capabilities and power. These metrics are crucial to determine the suitability and help in making informed decisions on choosing the right cryptographic algorithms to strike a balance between security and performance. Through the evaluation it is observed that SPECK exhibits better performance in resource-constrained IoT devices.

MỘT THUẬT TOÁN HỮU ÍCH ĐỂ KHAI THÁC TẬP HỮU ÍCH CAO

High utility itemsets (HUIs) mining is the finding of itemsets that satisfy a user-defined minimum utility threshold. Many successful studies in this field have been carried out, however they are all reliant on Tidset techniques, which records the intersection of transactions in a data structure. This paper presents the DCHUIM algorithm which mines the high utility itemset based on the Diffset techniques. Essentially, this mechanism stores the subtraction set of transactions rather than the intersection set. In order to achieve this, a DUL data structure is proposed to store utilities information and subtraction transactions of an itemset. Furthermore, the algorithm also applies pruning strategies such as U-Prune, EUCS-Prune and the concept of closed utility to effectively compress data. Thus, in the mining process, the search space is greatly diminished. Experiment on large datasets including Accidents, Mushroom, Retail, Chainstore and compare the performance of DCHUIM algorithm with HMiner algorithm. The findings indicate that the DCHUIM method outperforms the HMiner algorithm in terms of memory utilization across all databases and outperforms it in terms of time on sparse databases.

Efficient context-aware computing: a systematic model for dynamic working memory updates in context-aware computing.

The expanding computer landscape leads us toward ubiquitous computing, in which smart gadgets seamlessly provide intelligent services anytime, anywhere. Smartphones and other smart devices with multiple sensors are at the vanguard of this paradigm, enabling context-aware computing. Similar setups are also known as smart spaces. Context-aware systems, primarily deployed on mobile and other resource-constrained wearable devices, use a variety of implementation approaches. Rule-based reasoning, noted for its simplicity, is based on a collection of assertions in working memory and a set of rules that regulate decision-making. However, controlling working memory capacity efficiently is a key challenge, particularly in the context of resource-constrained systems. The paper's main focus lies in addressing the dynamic working memory challenge in memory-constrained devices by introducing a systematic method for content removal. The initiative intends to improve the creation of intelligent systems for resource-constrained devices, optimize memory utilization, and enhance context-aware computing.

COGNITIVE PROCESSING AND MEMORY TECHNIQUES IN CONSECUTIVE INTERPRETATION

This study investigates the psychophysiological characteristics and memory utilization in professional translators specializing in consecutive interpretation. Utilizing a comprehensive questionnaire, the study gathered data from practicing translators to understand how mechanical, logical, and imaginative memory strategies impact their translation performance, particularly focusing on spoken information retention over extended periods. Results indicate a strong preference for logical memorization due to its effectiveness in structuring and understanding complex information. However, imaginative memorization also plays a crucial role in scenarios requiring dynamic translation solutions. These findings suggest that training programs should emphasize versatile memory strategies to enhance translation accuracy and efficiency.

Advancing Supervised Learning with the Wave Loss Function: A Robust and Smooth Approach

Loss function plays a vital role in supervised learning frameworks. The selection of the appropriate loss function holds the potential to have a substantial impact on the proficiency attained by the acquired model. The training of supervised learning algorithms inherently adheres to predetermined loss functions during the optimization process. In this paper, we present a novel contribution to the realm of supervised machine learning: an asymmetric loss function named wave loss. It exhibits robustness against outliers, insensitivity to noise, boundedness, and a crucial smoothness property. Theoretically, we establish that the proposed wave loss function manifests the essential characteristic of being classification-calibrated. Leveraging this breakthrough, we incorporate the proposed wave loss function into the least squares setting of support vector machines (SVM) and twin support vector machines (TSVM), resulting in two robust and smooth models termed as Wave-SVM and Wave-TSVM, respectively. To address the optimization problem inherent in Wave-SVM, we utilize the adaptive moment estimation (Adam) algorithm, which confers multiple benefits, including the incorporation of adaptive learning rates, efficient memory utilization, and faster convergence during training. It is noteworthy that this paper marks the first instance of Adam’s application to solve an SVM model. Further, we devise an iterative algorithm to solve the optimization problems of Wave-TSVM. To empirically showcase the effectiveness of the proposed Wave-SVM and Wave-TSVM, we evaluate them on benchmark UCI and KEEL datasets (with and without feature noise) from diverse domains. Moreover, to exemplify the applicability of Wave-SVM in the biomedical domain, we evaluate it on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. The experimental outcomes unequivocally reveal the prowess of Wave-SVM and Wave-TSVM in achieving superior prediction accuracy against the baseline models. The source codes of the proposed models are publicly available at https://github.com/mtanveer1/Wave-SVM.

SPMSD: An Partitioning-Strategy for Parallel General Sparse Matrix-Matrix Multiplication on GPU

SpGEMM (General Sparse Matrix-Matrix Multiplication) is one of the kernels of an algebraic multi-grid method, graph algorithm, and solving linear equations. Due to the non-uniformity of some sparse matrices, the existing parallel SpGEMM algorithms suffer from load imbalance, lead to a decrease in computational efficiency. This paper proposes a new algorithm, SPMSD (SpGEMM Based on Minimum Standard Deviation). The algorithm is developed based on a hash table and partition strategy. First, the number of intermediate results in the matrix is divided into multiple blocks based on a new partition strategy to ensure the minimum standard deviation among blocks. Second, the input matrix is transformed according to the result of the partition strategy. Finally, SPMSD performs the parallel computing of SpGEMM based on the advantages of fast insertion and also fast access storage of the hash table and the calculation process controls the insertion and merging of intermediate results according to the offset to avoid the shortage of atomic operations. These experiments indicate the execution of SPMSD is faster than the existing cuSPARSE libraries by 7.4x. Compared with the Out of Core method, SPMSD improves the computational performance by 1.2x, SPMSD memory utilization is decreased by 0.19x.

GLSTM: On Using LSTM for Glucose Level Prediction.

The Prediabetes impacts one in every three individuals, with a 10% annual probability of transitioning to type 2 diabetes without lifestyle changes or medical interventions. It's crucial to manage glycemic health to deter the progression to type 2 diabetes. In the United States, 13% of individuals (18 years of age and older) have diabetes, while 34.5% meet the criteria for prediabetes. Diabetes mellitus and prediabetes are more common in older persons. Currently, nevertheless, there aren't many noninvasive, commercially accessible methods for tracking glycemic status to help with prediabetes self-management. This study tackles the task of forecasting glucose levels using personalized prediabetes data through the utilization of the Long Short-Term Memory (LSTM) model. Continuous monitoring of interstitial glucose levels, heart rate measurements, and dietary records spanning a week were collected for analysis. The efficacy of the proposed model has been assessed using evaluation metrics including Root Mean Square Error (RMSE), Mean Squared Error (MSE), Mean Absolute Error (MAE), and the coefficient of determination (R2).

Hardware Efficient Integrated In-loop Filter for HEVC Encoder

The deblocking filter (DF) and the sample adaptive offset (SAO) filter, which aids in enhancing the subjective quality of the image, make up the in-loop filter of the high-efficiency video coding (HEVC) encoder and decoder. The in-loop filter significantly increases the computational load on the HEVC encoder. It is challenging to design an in-loop filter on hardware that can handle intensive computations while using the least amount of on-chip memory, taking external memory traffic and dependencies simultaneously delivering high throughput to support Ultra HD video applications. The proposed design employs the following strategies to address these issues. This work proposes an address generation technique for pipelined horizontal and vertical filtering in DF, that avoids a transpose buffer which otherwise is required. This enables easy pipelining and parallelization thus improving throughput while reducing the on-chip memory utilization. A simplified SAO filter with parallel-pipelined processing is included in the design. These features enable the design to support ultra-HD 7680 × 4320 @ 40 fps video applications. The proposed hardware architecture has a total gate count of 7.73 K LUTs and 2.8 K slice registers, and it is implemented on a 28 nm field programmable gate array (FPGA) platform.

Predicting Peak Ground Acceleration of Strong-Motion Earthquakes Using Variable Snapshots of P-Wave Data with Long Short-Term Memory Neural Network

Abstract Conventional earthquake early warning systems (EEWS) rely on mathematical functions that utilize P-wave parameters extracted over a 3 s window to estimate peak ground acceleration (PGA). Advancements in the capabilities of deep neural networks to approximate universal functions, coupled with the availability of strong seismic event data, offer an unprecedented opportunity to evaluate the relationship between variable snapshots of P-wave data and the PGA of strong-motion earthquakes. This convergence of technology and data opens new avenues for research into utilizing smaller snapshots of P-wave in EEWS. Our study centers on the utilization of a long short-term memory (LSTM) neural network to model long dependencies within the P-wave of 1839 earthquakes recorded by the Kiyonshin Network (K-NET) for the prediction of PGA of S waves. Our methodology involves experiments that ultimately evaluate the network’s performance on 4, 3, and 2 s of P-wave snapshots. Our findings indicate that there is sufficient information in 2 s of temporal accelerometer readings after the onset of P waves to predict PGA accurately with an LSTM network.