INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026  
CoreWatchAI-Driven CPU/GPU PerformanceAnalyzer  
Karan N, Nimay N, Jeevanandan, Puneeth MS, Divyaprabha KN  
Department of Computer Science and Engineering PES University Bangalore, Karnataka, India  
DOI: https://doi.org/10.51583/IJLTEMAS.2026.150100088  
Received: 30 January 2026; Accepted: 04 February 2026; Published: 14 February 2026  
ABSTRACT  
CoreWatch is a lightweight, cross-platform monitoring system designed to provide real-time insights into CPU,  
GPU, memory, disk, and network performance. It uses Flask with Socket IO for low-latency metric streaming  
and integrates psutil and NVIDIA-SMI for accurate data collection. The dashboard employs Chart.js for  
smooth, interactive visualizations. An LSTM-based prediction module enhances monitoring by forecasting  
short-term CPU and GPU trends. CoreWatch maintains under 3% system overhead, ensuring efficient  
performance without additional load. Testing confirms stable responsiveness across Windows, Linux, and  
macOS. The system focuses on accessibility, minimal setup, and clear visual analysis. Future extensions include  
historical logging, improved AI models, and remote monitoring capabilities.  
Keywords: System monitoring, CPU–GPU performance analysis, LSTM prediction, anomaly detection,  
machine learning, real-time telemetry, resource optimization.  
INTRODUCTION  
Modern computing environments increasingly depend on effective resource monitoring, as system workloads  
grow more demanding across personal systems, professional applications, and cloud-based infrastructures.  
Gaining real-time visibility into CPU, GPU, memory, disk, and network performance has become crucial for  
maintaining system stability. Prior research highlights the importance of continuous monitoring and predictive  
analysis across heterogeneous networks [1], cloud systems [2], and real-time data applications [3], reinforcing  
the need for lightweight solutions that can adapt to changing workloads.  
Conventional monitoring tools such as Task Manager, Prometheus, or Grafana provide strong capabilities but  
often require heavy configuration, external databases, or persistent background services. These tools generally  
emphasize current readings rather than offering predictive insight. With recent progress in AI-driven  
optimization including CPU resource management [4], GPU behavior analysis [5], and time-series forecasting  
models [6] there is growing interest in platforms that combine intuitive visualization with intelligent trend  
prediction.  
CoreWatch is designed to address these needs by offering a cross-platform, browser-based dashboard that  
streams real-time system metrics along with short-term workload forecasts. The backend captures accurate  
CPU and GPU data using psutil and NVIDIA-SMI, while event-driven communication through Flask-SocketIO  
ensures smooth and continuous metric updates. The frontend, powered by Chart.js, presents these trends  
clearly, enabling both technical and non-technical users to interpret system behavior with ease. Additionally,  
CoreWatch integrates an LSTM-based prediction module that estimates upcoming CPU and GPU utilization,  
supporting proactive performance awareness similar to modern workload-forecasting methods used in  
heterogeneous computing environments to minimize performance failures and resource imbalance [7][10][11].  
To further improve usability, CoreWatch includes an integrated prediction layer that highlights short-term  
variations in CPU and GPU behavior. This capability provides early visibility into rising workloads and aligns  
with current developments in intelligent system management and edge-based resource forecasting. By  
combining responsive visualization with predictive analytics, the system delivers an intuitive pathway for users  
to understand and manage their device performance more effectively.  
Page 1074  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026  
In summary, CoreWatch brings together modern concepts in system monitoring by providing a lightweight,  
cross-platform dashboard capable of real-time CPU, GPU, memory, disk, and network analysis. Unlike  
traditional tools that require complex setup or focus only on current statistics, CoreWatch leverages  
WebSocket-based updates and accurate telemetry collection through psutil and NVIDIA-SMI. With its clean  
visualization layer and integrated LSTM prediction model, the system offers a practical, accessible, and  
intelligent solution for tracking system performance and anticipating workload spikes across diverse computing  
environments.  
LITERATURE SURVEY  
System monitoring technologies have undergone significant evolution, shifting from basic resource reporting  
tools to more intelligent, adaptive, and analytics-driven frameworks. Early monitoring utilities including  
Windows Task Manager and Linux top/htop provided essential visibility into CPU and memory usage but  
lacked cross-platform consistency, extensibility, and predictive intelligence. They remain limited to real-time  
snapshots without offering trend analysis or forecasting capabilities, which are increasingly required in modern,  
heterogeneous computing environments.  
Recent research emphasizes the need for scalable, real-time monitoring solutions capable of handling diverse  
architectures and dynamic workloads. Aldea et al. [1] proposed an integrated monitoring architecture for  
heterogeneous networks, highlighting the importance of unified frameworks but also revealing the complexity  
and overhead associated with such systems. A broad systematic review by da Costa et al. [3] further confirmed  
that most real-time monitoring solutions rely heavily on centralized data pipelines, making them unsuitable for  
lightweight or personal systems. These findings underline the demand for monitoring tools that offer both  
flexibility and low resource footprints.  
In parallel, the rise of cloud and HPC workloads has accelerated research on predictive analytics for resource  
utilization. Yildirim [2] and Patel & Bedi [7] demonstrated that machine learning models—including GRU,  
LSTM, and attention-based architectures—significantly enhance workload forecasting accuracy in cloud  
environments. Similarly, Nashold and Krishnan [6] explored LSTM and SARIMA models for CPU prediction  
in cluster systems, reinforcing the value of time-series modeling for proactive performance management. These  
studies collectively support the integration of ML-based forecasting in modern monitoring tools—an approach  
CoreWatch adopts for CPU prediction.  
GPU-aware monitoring has also expanded, especially with the rise of AI and edge computing. Woo [5]  
investigated GPU sharing mechanisms for smart city applications, revealing the need for fine-grained GPU  
telemetry in distributed systems. Weakley et al. [8] provided insights into GPU characterization across HPC  
workloads, showing how GPU metrics correlate strongly with performance bottlenecks. These observations  
justify CoreWatch’s GPU monitoring and visualization features via NVIDIA-SMI.  
Researchers have also explored visualization and interaction methods for performance analytics. Shilpika et al.  
[9] introduced a visual-analytics approach for hardware monitoring using streaming functional data,  
emphasizing clarity and interpretability. Additionally, Huang et al. [10] showed how edge-based monitoring  
can improve responsiveness and local decision-making, reinforcing the need for lightweight, on-device  
telemetry systems similar to CoreWatch’s architecture. Wang and Xing [4] further highlighted the role of AI-  
driven CPU resource management in modern operating systems, validating CoreWatch’s choice to include  
prediction capabilities.  
Despite these advancements, the literature reveals clear gaps in accessibility, low-overhead design, and user-  
centered interaction. Most existing tools require complex setup pipelines, high processing overhead, or lack  
predictive features. Few provide conversational interfaces or web-based, cross-platform dashboards that remain  
lightweight while supporting intelligent analytics. CoreWatch addresses these shortcomings by integrating real-  
time telemetry, AI-driven predictions, and a chatbot-driven interaction layer into a unified, resource-efficient  
system.  
Overall, the surveyed literature collectively underscores the need for a monitoring framework that combines  
Page 1075  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026  
real-time data acquisition, predictive modeling, intuitive visualization, and broad platform compatibility.  
CoreWatch emerges directly from these research insights, delivering an accessible, modular, and intelligent  
monitoring solution that bridges the gap between heavy enterprise frameworks and simple system utilities.  
Proposed Approach  
This section presents the design and operational methodology of the CoreWatch AI-Powered Monitoring  
Dashboard. The system architecture emphasizes intelligent, lightweight monitoring through modular  
integration of data acquisition, real-time communication, predictive analytics, and user interaction. First, the  
metric collection model is described. Second, the real-time communication and data transmission model are  
detailed. Third, the visualization, alerting, and AI-based prediction modules are presented. Finally, the  
optimization, chatbot integration, and overall workflow are discussed.  
Metric Collection Model  
The CoreWatch framework begins by acquiring live system metrics from the host machine using psutil and  
NVIDIA System Management Interface (nvidia-smi) libraries. Each monitored parameter—CPU, GPU,  
memory, disk, network, temperature, and battery, is represented as an independent metric Miwithin the global  
set.  
푀 = {푀1, 푀2, 푀3, … , 푀푛}  
Each metric 푀푖is sampled periodically at an interval Δ푡(default = 2 seconds), ensuring real-time responsiveness  
while maintaining minimal computational overhead. The data collection engine executes asynchronously using  
Python threads, allowing smooth performance even under heavy system load.  
GPU utilization, temperature, and memory usage are extracted via nvidia-smi in structured JSON format, while  
CPU and system-level data are obtained using psutil.  
All collected metrics are normalized into a unified JSON schema for consistent communication with the  
backend streaming layer.  
Real-Time Communication and Data Transmission Model  
CoreWatch uses Flask as the central backend engine, enhanced with Flask-SocketIO to enable real-time,  
bidirectional communication between the server and the browser-based dashboard. Unlike traditional REST  
polling where the client repeatedly requests fresh data CoreWatch adopts a continuous, event-driven  
WebSocket pipeline. This significantly reduces redundant traffic, lowers latency, and ensures smoother metric  
streaming under varying workloads.  
Let 푅푖(푡)denote the real-time reading of metric 푀푖at time 푡. The complete data stream is represented as:  
푆(푡) = {푅1(푡), 푅2(푡), … , 푅푛(푡)}  
This stream is continuously broadcast to all connected dashboard clients through SocketIO channels. The  
communication follows a publisher–subscriber architecture: the backend acts as the publisher, while the  
frontend clients subscribe to updates. This approach significantly reduces redundant HTTP requests, ensuring  
near-instant updates of all visual and analytical components while maintaining low network overhead.  
Visualization and Alert Generation Model  
The CoreWatch frontend is built using HTML, CSS, and Chart.js, offering an interactive and visually refined  
dashboard with separate pages for CPU, GPU, memory, network, and disk metrics. Chart.js dynamically  
redraws graphs upon receiving new SocketIO events, ensuring uninterrupted visualization without page  
refreshes.  
Page 1076  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026  
The alerting mechanism continuously compares current readings 푅푖(푡)with predefined threshold values 푇 . An  
푖
alert event 퐴푖(푡)is triggered as:  
1, if 푅푖(푡) ≥ 푇  
푖
퐴푖(푡) = {  
0,  
otherwise  
When triggered, alerts are logged and displayed with visual cues (color indicators, timestamps, and severity  
levels) on the dashboard.  
The prediction model evaluates short sequences of recent CPU and GPU readings (default sequence length L  
̂
̂
= 10) and estimates the next expected utilization values 푅푐푝푢(푡 + 1) and 푅푔푝푢(푡 + 1). If the trained LSTM  
model or its corresponding scaler is unavailable, CoreWatch automatically falls back to a lightweight linear  
estimation method to ensure uninterrupted predictive output. These forecasts enable early detection of potential  
performance spikes, allowing users to take proactive actions before resource thresholds are exceeded.  
Optimization and Performance Considerations  
CoreWatch is designed to remain efficient across a wide range of systems, from low-power laptops to high-  
performance workstations. Its architecture avoids heavy databases or logging services, instead using  
lightweight, in-memory buffers that refresh periodically. This reduces disk I/O and prevents unnecessary  
storage overhead.  
Furthermore, by using Flask-SocketIO with asynchronous background threads, the system maintains real-time  
responsiveness with an average CPU overhead consistently below 3%. The sampling interval Δ푡is adaptive,  
balancing responsiveness with energy efficiency. CoreWatch also ensures that predictive inference is executed  
only when sufficient historical data is available, reducing computational load associated with machine learning  
processes.  
These design optimizations collectively enable sustained performance even during long-duration monitoring  
sessions, making CoreWatch suitable for continuous use in personal, academic, and professional environments.  
Operational Workflow  
The complete operational workflow of the CoreWatch AI-Powered Monitoring Dashboard integrates metric  
acquisition, real-time communication, visualization, and prediction into a unified loop. The system begins with  
the initialization of the Flask backend and WebSocket communication channels, followed by continuous data  
acquisition and live metric broadcasting. Each cycle of operation ensures seamless updates and proactive  
performance awareness across all components using the following steps.  
1. Start  
2. Initialize Flask backend and establish WebSocket channels using Flask-SocketIO.  
3. Collect real-time system metrics 푀푖through psutil and NVIDIA-SMI APIs.  
4. Normalize and format the collected data into a unified JSON schema.  
5. Stream metric data 푆(푡)to connected dashboard clients via SocketIO.  
6. Visualize CPU, GPU, and other system parameters dynamically using Chart.js components.  
7. Compare each metric 푅푖(푡)with its corresponding threshold 푇 ; if 푅푖(푡) ≥ 푇 , trigger alert event 퐴푖(푡).  
푖
푖
8. Execute prediction module to estimate future CPU and GPU utilization using the LSTM model or fallback  
predictor.  
Page 1077  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026  
9. Log alert events, predictions, and system performance summaries for analysis.  
10. Repeat monitoring and update cycle at defined interval Δ푡.  
11. Stop  
This integrated workflow ensures low-latency, high-efficiency monitoring suitable for personal, educational,  
and professional environments.  
RESULTS AND DISCUSSION  
This section presents the implementation details and performance analysis of the CoreWatch AI-Powered  
Monitoring Dashboard, comparing it with traditional system monitoring tools. The framework was  
implemented using Python Flask, Flask-SocketIO, and psutil for backend data acquisition, with Chart.js for  
frontend visualization. The system was tested across Windows 11, Ubuntu 22.04, and macOS Ventura to  
validate cross-platform compatibility.  
Performance metrics considered for evaluation include CPU responsiveness, GPU monitoring accuracy, system  
overhead, and real-time predictive efficiency. The figures in this section represent actual dashboard snapshots  
and measured outcomes under varying system loads.  
CPU Utilization and Responsiveness  
CoreWatch efficiently captures and visualizes CPU utilization metrics in real time using the psutil library. Per-  
core utilization, average system load, and process counts are collected asynchronously every 2 seconds and  
transmitted via SocketIO to the frontend dashboard.  
The asynchronous data collection process runs in independent background threads, ensuring that even during  
heavy system activity, the monitoring engine remains non-blocking and stable. Each collected data point is  
instantly normalized into a structured JSON format and transmitted to the frontend dashboard via Flask-  
SocketIO channels, eliminating the need for frequent HTTP polling requests. This event-driven communication  
design enables smooth, uninterrupted data flow between backend and frontend components.  
Figure 1 illustrates the CPU Dashboard, showing real-time CPU usage graphs, average utilization, and process  
count updates. The chart dynamically refreshes without manual reload, maintaining a latency of less than 150  
milliseconds between backend sampling and frontend visualization.  
Figure 1. Real-time CPU Utilization Dashboard.  
Page 1078  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026  
Experimental testing demonstrated that Corewatch’s CPU monitoring module operates with less than 3%  
CPU overhead, significantly lower than traditional monitoring tools like Windows Task Manager or Netdata,  
which can consume between 5–8% CPU resources. This lightweight performance highlights the framework’s  
suitability for continuous monitoring.  
In addition to real-time display, CoreWatch’s CPU module also maintains short-term historical data buffers  
that allow users to visually analyze workload fluctuations and temporal patterns over time. This feature  
provides valuable insight into system behavior without requiring persistent database storage. Overall, the  
combination of low overhead, high responsiveness, and accurate visualization establishes CoreWatch as a  
highly effective solution for continuous system performance monitoring.  
GPU Monitoring and Temperature Tracking  
GPU monitoring in CoreWatch is implemented using the NVIDIA System Management Interface (nvidia-  
smi), which provides direct access to GPU hardware statistics through a structured JSON output. The  
monitoring module retrieves key parameters such as GPU utilization percentage, core temperature, and power  
consumption, ensuring comprehensive visibility into graphics hardware performance.  
Figure 2 presents the GPU Monitoring Dashboard, where the real-time GPU usage and thermal readings are  
plotted using Chart.js. The readings are refreshed at the same interval (2 seconds) to ensure synchronization  
with CPU data.  
Real-time GPU Utilization and Temperature Dashboard.  
The GPU monitoring module introduces minimal additional load, with average execution latency below 0.5  
seconds per sampling cycle. Comparative analysis shows that Corewatch’s data accuracy aligns closely with  
NVIDIA’s official desktop monitoring tools, confirming reliability. Additionally, the GPU module supports  
predictive workload estimation using short-term historical sequences processed by the integrated LSTM-based  
model, enabling CoreWatch to forecast upcoming spikes in GPU utilization during graphically intensive tasks  
such as gaming, rendering, or AI model inference. This predictive layer enhances the user’s ability to manage  
workloads proactively, preventing overheating and maintaining stable performance.  
Memory/Disk, and Network Performance  
In addition to CPU and GPU tracking, CoreWatch provides dedicated dashboards for monitoring memory, disk,  
and network performance, ensuring a complete overview of system health in real time.  
The memory module continuously measures and displays parameters such as total memory, used memory, and  
Page 1079  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026  
available memory, allowing users to easily identify potential memory bottlenecks or abnormal consumption  
patterns. Similarly, the disk activity dashboard provides live insights into overall disk utilization and  
performance trends, helping users access system load during intensive operations such as application launches  
or background data processing.  
Figure 3 illustrates the integrated Memory and Disk Utilization Dashboard, where each metric is represented  
through interactive and visually distinct charts designed for both clarity and accessibility.  
Figure 3. Memory and Disk Utilization Dashboard.  
The network monitoring module in CoreWatch provides a clear visualization of real-time incoming and  
outgoing bandwidth, enabling users to detect abnormal data transfer rates, excessive background activity, or  
potential network bottlenecks with ease. By continuously updating every two seconds, the module ensures that  
network usage patterns are accurately reflected, helping users maintain optimal connectivity and system  
performance.  
Figure 4 illustrates the Real-Time Network Activity Dashboard, showcasing dynamic bandwidth graphs that  
simplify the interpretation of network behavior for both technical and non-technical users.  
Figure 4. Real-time Network Activity Dashboard.  
Page 1080  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026  
Alert Mechanism and Intelligent Chatbot Interface  
The CoreWatch dashboard integrates a threshold-based alert system and an AI-powered chatbot interface,  
creating a unified, interactive, and user-centric monitoring experience. The alert module continuously tracks  
system parameters against configurable thresholds, automatically generating color-coded warning cards when  
limits are exceeded (e.g., CPU > 90%, GPU > 95%, temperature > 80 °C). These alerts appear in real time with  
severity indicators and timestamps, enabling users to respond promptly to performance issues. Complementing  
this, the AI chatbot allows users to query system performance through simple natural-language commands such  
as “Show GPU usage” or “What is my CPU temperature?”, interpreting user intent and providing real-time  
metrics with contextual recommendations. Together, the alert system and chatbot enhance accessibility and  
situational awareness for both technical and non-technical users, bridging the gap between traditional  
dashboards and conversational system management.  
Figures 5 and 6 illustrate the alert notification system and AI-based chatbot interface, respectively,  
demonstrating how CoreWatch simplifies system insights through an intelligent, visually intuitive design.  
Figure 5. Threshold-Based Alert Notification System  
Figure 6. AI-powered chatbot  
Page 1081  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026  
Together, the alert mechanism and chatbot interface transform CoreWatch from a conventional monitoring tool  
into an interactive, intelligent assistant. This integrated design ensures rapid response, improved usability, and  
a more intuitive monitoring experience overall.  
Comparative System Overhead Analysis  
To quantify efficiency, Corewatch’s performance was benchmarked against Grafana, Task Manager, and  
Netdata using identical test conditions on a 12th Gen Intel i5 system. Results show that Corewatch achieved  
up to 60% lower resource overhead while maintaining comparable data update frequency and visualization  
quality.  
Table 1. Comparative Performance Analysis of System Monitoring Tools  
This comparative evaluation confirms that CoreWatch achieves an optimal balance between performance,  
accessibility, and system efficiency. Its low overhead, cross-platform adaptability, and minimal setup  
requirements make it an ideal solution for personal, academic, and lightweight production environments.  
CONCLUSION  
The study identified that many existing system monitoring solutions, despite being feature-rich, are resource-  
heavy, difficult to configure, and often restricted to specific platforms making them unsuitable for everyday  
users, developers, and small-scale deployments. To overcome these limitations, this paper introduced  
CoreWatch, an AI-enabled, lightweight, and cross-platform monitoring framework designed to deliver real-  
time performance insights through a modular and efficient architecture. Leveraging Flask and Flask-SocketIO  
for low-latency data streaming, psutil and NVIDIA-SMI for accurate metric acquisition, and Chart.js for  
dynamic visualization, CoreWatch provides continuous monitoring with minimal resource consumption.  
Experimental testing across Windows, macOS, and Linux confirmed that CoreWatch sustains an average CPU  
overhead below 3%, outperforming tools such as Grafana and Netdata in terms of responsiveness and efficiency  
while delivering comparable visualization quality. The system reliably monitors CPU, GPU, memory, disk,  
and network performance, and incorporates intelligent features such as threshold-based alerting and a natural-  
language chatbot interface to enhance accessibility for both technical and non-technical users.  
In conclusion, CoreWatch emerges as a practical, scalable, and intelligent alternative to conventional  
monitoring platforms by combining professional-grade insight with lightweight deployment and simplicity of  
use. Future work aims to extend the framework with advanced machine learning models for deeper predictive  
analytics, cross-device remote monitoring capabilities, and database integration to support long-term historical  
trend analysis. These enhancements will help transform CoreWatch into a fully adaptive performance  
management ecosystem capable of proactive decision-making and smarter system optimization.  
Page 1082  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026  
REFERENCES  
1. C. L. Aldea, R. Bocu, and R. N. Solca, “Real-Time Monitoring and Management of Hardware and  
Software Resources in Heterogeneous Computer Networks through an Integrated System Architecture,”  
Symmetry, vol. 15, no. 6, article 1134, 2023.  
2. E. Yildirim, “Predicting Runtime and Resource Utilization of Jobs on Integrated Cloud and HPC  
Systems,” Future Generation Computer Systems, 2025.  
3. T. P. da Costa, S. A. de Oliveira, et al., “A Systematic Review of Real-Time Data Monitoring and Its  
Applications,” Expert Systems with Applications, vol. 242, article 123880, 2024.  
4. Y. Wang and S. Xing, “AI-Driven CPU Resource Management in Cloud Operating Systems,” Journal  
of Computer and Communications, vol. 13, pp. 135–149, 2025.  
5. S. Woo, “Exploring GPU Sharing Techniques for Edge AI Smart City Applications,” ETRI Journal,  
Wiley Online Library, vol. 47, no. 2, pp. 112–123, 2025.  
6. L. Nashold and R. Krishnan, “Using LSTM and SARIMA Models to Forecast Cluster CPU Usage,”  
arXiv preprint arXiv:2007.08092, 2020.  
7. Y. S. Patel and J. Bedi, “MAG-D: A Multivariate Attention and GRU Based Deep Learning Approach  
for Cloud Workload Forecasting,” Future Generation Computer Systems, vol. 142, pp. 376–392, 2023.  
8. L. M. Weakley, A. A. Sim, and S. Canon, “Monitoring and Characterizing GPU Usage,” in Proc. Cray  
User Group (CUG 2023), Helsinki, Finland, 2023.  
9. F. Shilpika, T. Fujiwara, N. Sakamoto, and H. Takemura, “A Visual Analytics Approach for Hardware  
System Monitoring with Streaming Functional Data Analysis,” arXiv preprint arXiv:2009.03488, 2020.  
10. J. Huang, S. Zhou, G. Li, et al., “Real-Time Monitoring and Optimization Methods for User-Side Energy  
Management Based on Edge Computing,” Scientific Reports, vol. 15, article 24890, Nature Publishing  
Group, 2025.  
11. M Divyaprabha Kabbal Narayana and Sudarshan Tekal Subramanyam Babu, “Optimal task partitioning  
to minimize failure in heterogeneous computational platform,” International Journal of Electrical and  
Computer Engineering, vol. 15, no. 1, pp. 1079–1088, Feb. 2025, doi: 10.11591/ijece.v15i1.pp1079-  
1088.  
Page 1083  
www.rsisinternational.org