Cloud data centers are essential for delivering scalable computing and storage services, yet challenges such as inefficient resource utilization, workload imbalance, and high energy consumption continue to impact performance and operational costs. This paper proposes a dynamic virtualization technique to optimize data server utilization in cloud environments by integrating real-time workload monitoring, adaptive virtual machine (VM) allocation, intelligent resource scheduling, and dynamic VM migration. The framework incorporates auto-scaling mechanisms to manage workload fluctuations, reducing idle resources while preventing server overload, and includes a predictive workload analysis component to forecast demand and allocate resources proactively. The proposed system is evaluated using performance metrics such as CPU utilization, memory efficiency, response time, throughput, and energy consumption. Experimental results demonstrate that the dynamic virtualization approach significantly improves server utilization, reduces power consumption, and enhances overall system performance compared to traditional static resource allocation methods, thereby supporting scalable, cost-effective, and sustainable cloud data center management

Attention-based workload prediction and dynamic resource allocation for heterogeneous computing environments.

An online algorithm for power consumption prediction of HPC workload

Network Function Virtualization (NFV) is a traditional network method, it’s replacing rigid, dedicated hardware devices with flexible, software-based virtual network functions that can be dynamically assigned and scaled across any networking infrastructure. Even though these findings give promise, still in the early stages of creating network management systems that are truly dependable, can handle massive scale, and keep sensitive information protected across different networking boundaries. While this sounds great in theory, there is a problem with how we currently use artificial intelligence to manage these systems. Traditional AI approaches, specifically reinforcement learning models, hit several problems. However, these methods often provide high computational overhead, slow convergence, and suffer from limited interpretability. In this paper, we propose a novel framework that replaces DRL with lightweight and interpretable Machine Learning (ML) algorithms. This article suggests a Federated Forcart-based NFV Resource Allocation Framework that combines a hybrid Random Forest–Cart prediction model with federated learning to address these issues. The framework supports balanced multi-metric resource optimization, precise workload prediction, and distributed knowledge exchange while maintaining local data secrecy. It maintains the parallelization strategy to minimize end-to-end latency, while significantly reducing training complexity and communication overhead. Improved prediction accuracy, decreased latency, and increased CPU and energy efficiency are demonstrated by comparison with state-of-theart methods. For next-generation NFV orchestration, the suggested method offers a scalable and private solution.

The rapid advancement of cloud computing has enabled large-scale deployment of machine learning applications; however, the emergence of Quantum Machine Learning (QML) introduces new challenges for cloud workload management. QML workloads exhibit unique execution patterns, hybrid quantum–classical behavior, and distinct resource requirements that are not effectively handled by existing cloud workload prediction and scheduling mechanisms. This paper presents a comprehensive review of recent research on cloud workload optimization, scheduling, and characterization, highlighting their limitations in supporting QML workloads. Based on the identified research gaps, a system model and methodology for QML workload prediction and categorization in cloud computing environments are discussed. The proposed approach emphasizes early workload analysis, feature-based prediction, and intelligent categorization to enable proactive scheduling and efficient resource utilization. Furthermore, key challenges related to quantum hardware limitations, hybrid integration complexity, and prediction reliability are analyzed. This study provides foundational insights for designing scalable and intelligent cloud platforms capable of supporting next-generation Quantum Machine Learning applications

This research explores the use of physiological signals derived from heart activity to assess mental effort during flight-related tasks. Data were collected through wearable sensors during simulations with varying cognitive demands. Specific indicators related to heart rate variability (HRV) were extracted and tested in different combinations to identify those most relevant for distinguishing levels of mental workload (WL). A Random Forest (RF) ensemble method is applied to classify two conditions, and its performance is examined under various settings, including model complexity and data partitioning strategies. Results showed that certain feature pairs significantly enhanced classification accuracy. The best features settings obtained from the RF are then used to train the other two decision trees-based classifiers, namely the AdaBoost and the XGBoost. Moreover, the decision trees models output is compared with predictions from a Kriging spatial interpolation technique, showing superior results in terms of reliability and consistency. This study highlights the potential of using heart-based physiological data and advanced classification techniques for developing intelligent support systems in aviation.

The intelligent computing infrastructures based on artificial intelligence (AI) have substantially increased the energy usage, lag times in computation, and costs of sustainability due to the exponential rise in the workloads. The classical models of workload management put much emphasis on predictive accuracy but ignore resource-awareness, with the net effect of inefficient usage of power and poor system responsiveness. This paper puts forward an Energy-Aware Hybrid CNN-LSTM-Transformer (EA-HCLT) architecture that would allow sustainable computing through the combination of workload prediction, smart scheduling and adaptive carving of model pruning to dynamic environments. The framework utilises workload prediction using hybrid learning/deep learning in real-time resource monitoring to optimally place computers to execute computations and also optimally use energy at maximum precision. As a validation of the effectiveness, EA-HCLT is compared to two popular models: Random Forest Workload Predictor (RF-WP) and Standard LSTM Scheduler (S-LSTM) based on the usage of synthetic workload and runtime workload datasets in terms of CPU, memory, network throughput, and accelerator utilisation. The overall analysis of the proposed approach in terms of accuracy, RMSE, latency, energy usage, sustainability index, and multi-objective cost reveal that the proposed solution provides a considerable improvement, yielding 14.8 percentage points higher accuracy, 19% reduced decision latency, 26.9% decreased energy usage, 17.5% higher sustainability index as opposed to S-LSTM. The results justify the supportability and scalability of the suggested EA-HCLT design and emphasise the significance of energy-conscious strategies of the next generation smart systems that are going to work within environmental and resource constraints.

The human systems literature is littered with conceptual models of human-automation interaction presented as a basis for understanding and explaining human performance effects. Unfortunately, the discrete and ordinal nature of existing models of levels of automation limits reliable prediction of operator performance, workload and situation awareness (SA). This paper presents enhanced quantitative models for determination of system automation proportion (AP) and SA in human-in-the-loop systems, building on an earlier preliminary model. We refine the AP concept as a continuous measure of level of system automation and introduce a generalised SA function that accounts for operator characteristics. The AP is calculated using hierarchical task analysis according to information processing stages. An overall proportion is then calculated for the system. The practicality and feasibility of this model are verified through a case study. We further propose a relationship between the AP and operator SA responses, based on existing empirical research findings.

Large-scale cloud data infrastructure faces significant challenges in balancing performance requirements with operational costs. This review examines dynamic capacity optimization (DCO) strategies and cost reduction mechanisms employed in modern cloud computing environments. The exponential growth of data volumes and computational demands has necessitated sophisticated approaches to resource allocation and workload management. Machine learning (ML) techniques have emerged as powerful tools for predictive capacity planning and automated resource provisioning. This paper synthesizes recent advances in workload prediction, autoscaling mechanisms, container orchestration, and energy-efficient computing architectures. We analyze various optimization algorithms including reinforcement learning (RL), genetic algorithms (GA), and hybrid approaches that combine multiple methodologies. The review also examines cost models, pricing strategies, and economic frameworks for cloud resource management. Particular attention is given to emerging technologies such as serverless computing, edge computing integration, and sustainable infrastructure designs. Through comprehensive analysis of contemporary research, this paper identifies key trends, technological innovations, and future directions in cloud capacity optimization. The findings indicate that integrated approaches combining predictive analytics, intelligent automation, and multi-objective optimization deliver superior results compared to traditional static provisioning methods. This review provides valuable insights for researchers and practitioners seeking to implement cost-effective, high-performance cloud infrastructure solutions.

This paper addresses the challenge of optimizing cloudlet resource allocation in a code evaluation system. The study models the relationship between system load and response time when users submit code to an online code-evaluation platform, LambdaChecker, which operates a cloudlet-based processing pipeline. The pipeline includes code correctness checks, static analysis, and design-pattern detection using a local Large Language Model (LLM). To optimize the system, we develop a mathematical model and apply it to the LambdaChecker resource management problem. The proposed approach is evaluated using both simulations and real contest data, with a focus on improvements in average response time, resource utilization efficiency, and user satisfaction. The results indicate that adaptive scheduling and workload prediction effectively reduce waiting times without substantially increasing operational costs. Overall, the study suggests that systematic cloudlet optimization can enhance the educational value of automated code evaluation systems by improving responsiveness while preserving sustainable resource usage.

Cloud data centers consume massive energy as workloads continue to grow. This study explores an energy-aware resource allocation framework that combines workload prediction and dynamic VM consolidation to reduce power consumption without compromising performance. The model uses historical utilization patterns to forecast demand and allocates resources accordingly. Experimental results on real datasets show noticeable reductions in energy use and SLA violations. The work contributes to sustainable cloud computing by balancing efficiency and reliability.

Real-time IoT applications such as smart surveillance and autonomous monitoring require low-latency, high-reliability cloud communication. Standard cloud schedulers often fail to accommodate dynamic IoT workloads. This study proposes an AI-enhanced cloud scheduling algorithm that integrates reinforcement learning with predictive workload analytics to reduce latency and maintain service stability. The scheduler evaluates device mobility patterns, network congestion, message frequency and compute demand to assign tasks optimally across edge and cloud nodes. Experiments were performed using the EdgeCloudSim and iFogSim platforms, simulating smart transportation and industrial IoT scenarios. The model achieved a 35 percent decrease in end-to-end latency and a 22 percent increase in throughput when compared with round-robin and priority-based schedulers. Results indicate that integrating ML into scheduling algorithms significantly enhances the responsiveness of IoT-cloud ecosystems. The method is scalable to multi-region cloud clusters and suitable for production deployment. Future work will explore multi-agent scheduling strategies.

This article will discuss the differences between microservice and monolithic architecture in real-time distributed architecture. Comparison cuts across theoretical underpinnings, performance attributes, development life cycles, operational issues, and implementation issues. Monolithic architectures are known to be beneficial in terms of simplicity, reduced baseline latency, and reduced communication overheads, and are applicable in applications with predictable workloads and intricate transactional requirements. Microservices, on the other hand, are much more scalable, fault-isolated, and focused on the allocation of resources, especially helpful in systems whose demands are variable and whose functionality evolves. The article discusses the issue of data consistency, overhead in inter-service communication, and state management complexities of distributed architectures, and notes pragmatic hybrid solutions and evolutionary trends that leverage the merits of each paradigm. The choice of architecture is actually determined by a set of constraints of particular projects, the structure of the organization, and the needs of real-time processing, instead of a particular architectural philosophy.

Abstract Accurately predicting resource load in cloud computing environments constitutes a fundamental challenge for dynamic resource allocation. Traditional threshold-based static scheduling strategies and linear time-series prediction methods struggle to address the nonlinear, abrupt changes and multi-time scale characteristics inherent in cloud workloads. Furthermore, existing deep learning approaches exhibit limitations in terms of noise robustness and multi-scale feature modeling. To overcome these challenges, this study introduces a novel Contrastive Learning Long Short-Term Memory (LSTM) Network model, termed SWT-CLSTM, which integrates Savitzky-Golay (SG) filtering with Smooth Wavelet Transform (SWT). This approach employs SG filters to preprocess and attenuate high-frequency noise, and utilizes SWT for the multi-resolution decomposition of low-frequency trends and high-frequency fluctuations. Additionally, the model incorporates a dual-path neural network architecture, comprising a one-dimensional Convolutional Neural Network (CNN) and an attention-enhanced LSTM. This architecture is designed to extract local patterns and model long-term dependencies. Moreover, the introduction of a frequency-aware hierarchical contrastive learning framework significantly enhances the model’s generalization capabilities for non-stationary data. Experimental evaluations conducted on public cloud task datasets confirm that the SWT-CLSTM model outperforms traditional methods and prevailing deep learning models across various time granularities, thereby markedly enhancing the temporal prediction accuracy of cloud computing resource scheduling.

The paper is about the problems of optimally scheduling the resources of GPUs when it comes to large-scale deep learning workloads within the cloud infrastructure. Due to the fact that the field of deep learning requires a noticeably large amount of computational power, GPUs play a crucial role in hastening these operations. Nevertheless, the management of the GPU resources in the cloud is still complicated because of the dynamics of the workload and resources allocation that requires efficient implementation. This paper introduces a predictive graphic card scheduling system, which takes advantage of machine learning to predict resource demands through the nature of workload. The architecture combines workload prediction, the control of the GPU resources, and optimization algorithms to allocate resources in advance that the deep learning tasks get the required amount of GIS in time and at the same time, the active time is minimized as well as the contention of the resources The framework relies on the historical performance data as well as analysis of workload behavior to forecast future demands which are then adjusted in terms of scheduling strategies. The strategy not only maximizes the use of the GPU but also leads to the overall performance of cloud-based systems since it minimizes the waste of resources and shortens the duration of tasks. To gauge the effectiveness of the suggested framework, the article logs the experiments carried on an extensive scale which shows the superiority of the suggested framework in comparison to the traditional ones in a large-scale cloud setting

Cloud computing has become the backbone of digital ecosystems, but growing workloads intensify challenges in resource optimization, virtual machine (VM) migration, and security assurance. Existing studies often address these issues in isolation, limiting their practical applicability. This paper presents a unified framework that integrates three complementary components: (i) an Improved Modified Particle Swarm Optimization (IMPSO) algorithm with adaptive inertia scheduling and dynamic mutation control, which outperforms IPSO in convergence speed and load distribution accuracy; (ii) a machine learning–assisted hybrid live VM migration method with dirty-page clustering and workload prediction to minimize downtime; and (iii) a blockchain-enabled secure migration layer to ensure tamper-proof and auditable state transfer. The revised version of this study includes statistical validation (confidence intervals, t-tests) and attack simulation experiments (e.g., man-in-the-middle and replay attacks) to ensure methodological rigor and realistic security assessment. Experimental results on a real XenServer testbed show that the proposed system improves response time by ~30%, reduces migration downtime by ~60%, and ensures 100% migration integrity with ?15% security overhead. Overall, this work represents among the first unified frameworks that jointly optimize resource allocation, downtime reduction, and blockchain-based security in a practically validated, end-to-end cloud migration environment.

PINE: Local patch reweighting and mixed independent neural encoder for datacentre workload prediction

Multitasking in mixed reality (MR) environments introduces unique cognitive demands, particularly in workload management. Accurate workload prediction is critical for optimizing user experience, safety, and performance in such settings. This study proposes a novel framework that integrates large language models (LLMs) with traditional workload assessment tools to enhance prediction accuracy in MR multitasking scenarios. A multitasking experiment involving 36 participants was conducted, combining real-world and virtual tasks, with workload evaluated using NASA-TLX. To address limited sample sizes, synthetic data was generated using generative adversarial networks (GANs), enabling robust model training. We employed a hybrid deep learning model that integrates LLM-generated text embeddings with numerical features in a feedforward neural network (FNN). Results show that integrating LLMs, specifically BERT and GPT-2, significantly improves workload prediction accuracy, with a root mean square error (RMSE) reduction from 6.82 (FNN-only) to 0.95 (BERT-integrated model). The findings underscore the potential of LLMs to augment cognitive workload assessment, supporting more adaptive and scalable human-machine collaboration in MR environments.

With the advent of big-scale smart computing, computational loads are growing exponentially, which has posed a danger to sustainability and scalability due to an increase in energy consumption. To solve this problem, the present paper proposes an energy-aware machine learning (ML) framework that can optimize its performance and reduce its power consumption in the distributed context. The framework incorporates deep learning (DL) models with energy-conscious scheduling and model pruning based on the heterogeneous datasets, such as CPU usage, memory usage, network usage, and system-energy information. The proposed system has adaptive learning mechanisms, unlike traditional approaches, which focus on the accuracy of predictions with no attention to the overhead of the resource allocation, which dynamically re-calibrates resource allocation according to the variations in the workload, enhancing the efficiency and resilience of the system. The algorithm is a hybrid CNN-LSTM workload prediction model with Transformer-based models to address long-term relations and use energy indicators in decision-making cycles. System performance-measured in predictive accuracy, decision latency, energy efficiency and sustainability index is mathematically modeled and optimized in the framework. Comparison of simulation-based predictive control proves the proposed approach to be 14.8 percent more predictive control-wise accurate, 27 percent energy consumption-wise less, and 19 percent latency-wise lesser than baselines like the Random Forest and standard LSTM models. Moreover, the stress tests at the peak loads and system volatility verify that the framework maintains a high level of adaptability, and the traditional approaches decline considerably. The proposed system illustrates how the energy-conscious ML can transform how decisions are supported through energy-efficient and accurate and scalable decision support. This study is a foundation of sustainable intelligent computation that represents the future of large-scale computing systems with an appropriate balance between performance and environmental responsibility.

Abstract In today's healthcare environments, especially for ones that are motivated by digital transformation and remote patient engagement, scalability of infrastructure is key to providing predictable access to services like Electronic Health Records (EHR), telemedicine platforms, and analytics dashboards for health. Classic auto-scaling of cloud environments like Azure Kubernetes Service (AKS) relies heavily on reactive thresholds mostly CPU or memory utilization to scale in terms of provisioning/deprovisioning containers. These approaches typically struggle with unpredictable traffic of high variance characteristic in healthcare (e.g., telehealth spikes in flu season or pandemics). This paper outlines an intelligent auto-scaling architecture with AI-driven predictive models specifically Long Short-Term Memory (LSTM) networks and Facebook’s Prophet to predict system demand. These models process past usage data from a simulated telehealth application and automatically invoke Kubernetes-based Event-Driven Autoscaler (KEDA) for adaptive management of pods. We have a real-case study scenario that emulates traffic surges in a digital health application and tests the system against conventional Horizontal Pod Autoscaler (HPA) approaches. Our results demonstrate improved response latency of 35% and cloud compute cost savings of 22% while sustaining 100% uptime. Our architecture met this twin need for resilience and cost-effectiveness in high-priority healthcare infrastructure and positions it as a building block for intelligent, scalable healthcare cloud solutions. Keywords Azure Kubernetes Service (AKS), Auto-scaling in cloud computing, Predictive scaling models, Healthcare cloud infrastructure, LSTM workload prediction, Prophet time-series forecasting, KEDA (Kubernetes Event-Driven Autoscaler), Telehealth optimization, AI in healthcare operations, Cloud-native architecture, HIPAA-compliant cloud solutions, MLOps in cloud scaling, Latency reduction in healthcare IT, Cost-efficient cloud resource management, Real-time digital health platforms