Deep neural networks in the cloud: Review, applications, challenges and research directions

Neurocomputing, Год журнала: 2023, Номер 545, С. 126327 - 126327

Опубликована: Май 15, 2023

Deep neural networks (DNNs) are currently being deployed as machine learning technology in a wide range of important real-world applications. DNNs consist huge number parameters that require millions floating-point operations (FLOPs) to be executed both and prediction modes. A more effective method is implement cloud computing system equipped with centralized servers data storage sub-systems high-speed high-performance capabilities. This paper presents an up-to-date survey on current state-of-the-art for computing. Various DNN complexities associated different architectures presented discussed alongside the necessities using We also present extensive overview platforms deployment discuss them detail. Moreover, applications already systems reviewed demonstrate advantages DNNs. The emphasizes challenges deploying provides guidance enhancing new deployments.

Язык: Английский

---

Advances and Predictions in Predictive Auto-Scaling and Maintenance Algorithms for Cloud Computing

Опубликована: Дек. 11, 2023

The use of sophisticated algorithms has radically altered cloud computing predictive auto-scaling and upkeep approaches. Recurrent Neural Networks (RNNs), the Prophet Algorithm, K-Means Clustering, Seasonal Autoregressive Integrated Moving-Average (SARIMA) all play a role in improving infrastructures, their interactions are studied here. By capitalizing on superiority processing sequential data, RNNs can deduce accurate workload forecasts from past patterns. Concurrently, Algorithm records seasonal annual patterns, which adds depth to forecasts. grouping servers into clusters with similar consumption Clustering improves resource allocation efficiency paves way for precise auto-scaling. SARIMA models capture nuanced fluctuations, lead reliable demand This explores state-of-the-art future directions these techniques, illuminating potential revolutionize current approaches management. When methods combined, service providers better able proactively scale resources, hence reducing likelihood bottlenecks outages. It foresees development subsequent widespread variety fields outside computing, such as Internet Things (IoT) networks edge infrastructures.

Язык: Английский

---

Cloud computing load prediction by decomposition reinforced attention long short-term memory network optimized by modified particle swarm optimization algorithm

Annals of Operations Research, Год журнала: 2023, Номер unknown

Опубликована: Дек. 15, 2023

Язык: Английский

---

An efficient proactive VM consolidation technique with improved LSTM network in a cloud environment

Computing, Год журнала: 2023, Номер 106(1), С. 1 - 28

Опубликована: Авг. 20, 2023

Язык: Английский

---

EvoGWP: Predicting Long-Term Changes in Cloud Workloads Using Deep Graph-Evolution Learning

IEEE Transactions on Parallel and Distributed Systems, Год журнала: 2024, Номер 35(3), С. 499 - 516

Опубликована: Янв. 23, 2024

Workload prediction plays a crucial role in resource management of large scale cloud datacenters. Although quite number methods/algorithms have been proposed, long-term changes not explicitly identified and considered. Due to shifty user demands, workload re-locations, or other reasons, the ”resource usage pattern” workload, which is usually stable short-term view, may change dynamically range. Such dynamic cause significant accuracy degradation for algorithms. How handle such an open challenging issue. In this paper, we propose Evolution Graph Prediction (EvoGWP), novel method that can predict using delicately designed graph-based evolution learning algorithm. EvoGWP automatically extracts shapelets identify patterns workloads fine-grained level, predicts by considering factors both temporal spatial dimensions. We design two-level importance based shapelet extraction mechanism mine new pattern dimension, graph model fuse interference among different dimension. By combining from each single workloads, then spatio-temporal GNN-based encoder-decoder workloads. Experiments real trace data Alibaba, Tencent Google show improves up 58.6% over state-of-the-art methods. Moreover, outperform methods terms convergence. To best our knowledge, first work identifies accurately

Язык: Английский

---

Enhanced virtual machine migration for energy sustainability optimization in cloud computing through knowledge acquisition

Computers & Electrical Engineering, Год журнала: 2024, Номер 119, С. 109506 - 109506

Опубликована: Июль 26, 2024

Cloud computing has revolutionized the way businesses and organizations manage their computational workloads. However, massive data centers that support cloud services consume a lot of energy, making energy sustainability critical concern. To address this challenge, article introduces an innovative approach to optimize consumption in environments through knowledge acquisition. The proposed method uses Knowledge Acquisition version Gray Wolf Optimizer (KAGWO) algorithm collect on availability use renewable within centers, contributing improved computing. KAGWO is introduced provide systematic for addressing complex problems by integrating global optimization principles, enhancing decision-making processes with fewer configuration parameters. This conducts comparative analysis between Swarm Intelligence Approach (KASIA) Genetic Algorithm (Pittsburgh) highlight benefits advantages former. By comparing performance KAGWO, Pittsburgh KASIA terms sustainability, study offers valuable insights into effectiveness knowledge-acquisition-based algorithms optimizing usage environments. results demonstrate outperforms offering more accurate acquisition capabilities, resulting enhanced sustainability. Overall, demonstrates substantial improvements ranging from 0.53% 5.23% over previous paper baselines, particular significance found slightly outperforming new small, medium large scenarios.

Язык: Английский

---

A Survey of Cloud Resource Consumption Optimization Methods

Journal of Grid Computing, Год журнала: 2025, Номер 23(1)

Опубликована: Янв. 8, 2025

Язык: Английский

---

Multilayer multivariate forecasting network for precise resource utilization prediction in edge data centers

Future Generation Computer Systems, Год журнала: 2025, Номер 166, С. 107692 - 107692

Опубликована: Янв. 10, 2025

Язык: Английский

---

A hybrid neural network and cooperative PSO model for dynamic cloud workloads prediction

Computing, Год журнала: 2025, Номер 107(3)

Опубликована: Фев. 15, 2025

Язык: Английский

---

Workload Prediction in Cloud Data Centers Using Complex‐Valued Spatio‐Temporal Graph Convolutional Neural Network Optimized With Gazelle Optimization Algorithm

Transactions on Emerging Telecommunications Technologies, Год журнала: 2025, Номер 36(3)

Опубликована: Март 1, 2025

ABSTRACT Workload prediction is the necessary factor in cloud data center for maintaining elasticity and scalability of resources. However, accuracy workload very low, because redundancy, noise, low center. In this manuscript, Prediction Cloud Data Centers using Complex‐Valued Spatio‐Temporal Graph Convolutional Neural Network Optimized with Gazelle Optimization Algorithm (CVSTGCN‐WLP‐CDC) proposed. Initially, input collected from two standard datasets such as NASA Saskatchewan HTTP traces dataset. Then, preprocessing Multi‐Window Savitzky–Golay Filter (MWSGF) used to remove noise redundant data. The preprocessed fed CVSTGCN a dynamic environment. work, proposed Approach (GOA) enhance weight bias parameters. CVSTGCN‐WLP‐CDC technique executed efficacy based on structure evaluated several performances metrics accuracy, recall, precision, energy consumption correlation coefficient, sum index (SEI), root mean square error (RMSE), squared (MPE), percentage (PER). provides 23.32%, 28.53% 24.65% higher accuracy; 22.34%, 25.62%, 22.84% lower when comparing existing methods Artificial Intelligence augmented evolutionary approach espoused centres architecture (TCNN‐CDC‐WLP), Performance analysis machine learning centered techniques (PA‐BPNN‐CWPC), Machine effectual utilization centers (ARNN‐EU‐CDC) respectively.

Язык: Английский

---