A novel approach to reliable and flexible distributed computing with virtualization in smart healthcare applications.

A novel approach to reliable and flexible distributed computing with virtualization in smart healthcare applications - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Sci Rep . 2026 Mar 5;16:12325. doi: 10.1038/s41598-026-40801-2 Search in PMC Search in PubMed View in NLM Catalog Add to search A novel approach to reliable and flexible distributed computing with virtualization in smart healthcare applications Gaurav Dhiman Gaurav Dhiman 1 Department of Computer Science and Engineering, Yuan Ze University, Tao Yuan, Taiwan 6 Centre of Research Impact and Outcome, Chitkara University, Rajpura, 140417, Punjab, India Find articles by Gaurav Dhiman 1, 6, ✉ , Kiran Deep Singh Kiran Deep Singh 2 Chitkara University, Rajpura, India Find articles by Kiran Deep Singh 2 , Prabh Deep Singh Prabh Deep Singh 3 Graphic Era Deemed to be University, Dehradun, Uttarakhand India Find articles by Prabh Deep Singh 3 , Norah Saleh Alghamdi Norah Saleh Alghamdi 4 Department of Computer Sciences, College of Computer and Information Sciences, Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia Find articles by Norah Saleh Alghamdi 4 , Ghadah Shukri Albakri Ghadah Shukri Albakri 5 Department of Teaching and Learning, College of Education and Human Development, Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia Find articles by Ghadah Shukri Albakri 5 Author information Article notes Copyright and License information 1 Department of Computer Science and Engineering, Yuan Ze University, Tao Yuan, Taiwan 2 Chitkara University, Rajpura, India 3 Graphic Era Deemed to be University, Dehradun, Uttarakhand India 4 Department of Computer Sciences, College of Computer and Information Sciences, Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia 5 Department of Teaching and Learning, College of Education and Human Development, Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia 6 Centre of Research Impact and Outcome, Chitkara University, Rajpura, 140417, Punjab, India ✉ Corresponding author. Received 2025 Oct 17; Accepted 2026 Feb 16; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13079764  PMID: 41786875 Abstract Currently, fast and efficient computing is driving the new disruptive technologies required by modern healthcare systems. A simulation model is presented that studies how virtualization affects the performance of task offloading in smart health-care environments. Virtualization technology is suitable for task-offloading processes. A simulation-based framework can be used to examine how virtualization overhead influences task offloading efficiency in smart healthcare environments. Therefore, the proposed model uses virtualization to manage resources more effectively and make the system more reliable. Important factors such as how long a task takes to finish, how much energy it uses, how much data it can handle, and how well it can grow when more tasks are added must be considered for smart healthcare. The results were compared with those of a system that did not use virtualization; thus, we can clearly observe how virtualization changes the overall performance. The findings show that better resource use, lower energy consumption, and improved fault recovery are possible, which makes the proposed system suitable for real healthcare applications. In the future, this work can be extended by adding predictive analytics, improving machine-learning–based scheduling, using multiple cloud platforms, and making virtualization even more energy-efficient. Keywords: Healthcare, Distributed, Fault tolerance, Virtualization Subject terms: Engineering, Health care, Mathematics and computing Introduction Distributed computing as a concept aims to transform the traditional definition of a computational system by allowing the sharing of physical resources, the distribution of tasks, and extensibility to other computers within a network 1 . As the demands of modern applications continue to grow, particularly in real-time data processing, big data analysis, and IoT, the need for achieving highly efficient, near-decentralised systems has become more critical than ever. Among the many advances that have driven the evolution, virtualization stands out as one of the key enablers, significantly enhancing the flexibility, reliability, and resource utilisation efficiency of distributed systems 2 . The paper examines the role virtualization techniques play in providing dependable and versatile distributed computing in edge-enriched networks. Distributed computing systems work in such a way that significant problems are solved by smaller sub-problems, which are solved by devices that are connected to a network 3 . Prior approaches to scheduling incorporated predetermined resource control, assigning activities to concrete nodes without considering the variability of workloads or the states of available resources. However, some applications where job distribution was pretty consistent and their processing did not vary with time, load, node failures, or any resource limitations 4 . In contrast, virtualization arises, creating an additional layer that isolates the hardware resources from the routines to be performed. In addition, virtualization entails the development of an electronic copy for physical bases; thus, leading to improved flexibility, sharing, and fault tolerance 5 . The flexibility is particularly relevant to edge-enriched networks, where computational assets can be located in different zones, and the loads they process may fluctuate over time. Due to these networks, virtualization is employed to manage resources and ensure that tasks are assigned to the most appropriate nodes, given the prevailing system conditions 6 – 10 . Resource utilisation is one of the benefits of virtualization in distributed computing. In traditional computing systems, resources are allocated fixedly to various tasks and/or nodes, resulting in situations where some nodes become overloaded with work. In contrast, other nodes are unoccupied 11 . Instead, virtualization enables the formation of resource pools and their allocation to various tasks or users. The dynamic give-and-take approach aids in the proper management of resources, minimising resource inactivity and maximising throughput. Besides, virtualization supports load balancing, whereby workloads are evenly shared across the nodes of a program to prevent previously exploited nodes from becoming overloaded. The third attribute of distributed systems, which is also improved by using virtualization 12 – 15 . In the conventional models, any node or component that fails to complete its duty will cause the entire system to shut down, or even result in data loss 16 . The risk is managed by virtualization since the failover mechanisms and redundancy are achievable in the environment. For example, when one of the nodes in a virtualised climate has failed, several tasks can be moved to other nodes without disturbing the entire operating system. The fault tolerance is especially advantageous in edge-enriched networks because nodes may vary in reliability due to power surges, connectivity disruptions, or even node failures 17 – 20 . The flexibility offered by virtualization is not limited to resource and fault tolerance flexibility. It also allows for the interconnection of dissimilar devices and systems present in a networked world. On one hand, edge-enriched networks are characterised by nodes, which may significantly differ in terms of hardware and software: the nodes’ OS and configuration. These different resources are all separate parts, but they can work together through a virtualized framework that lets them all talk to each other easily. In contrast, these problems are well solved by virtualization, which allows for optimised energy-conscious resource allocation. For instance, underutilised nodes can be powered down or grouped in a way that would minimise energy use without degrading actual performance. The capability is desirable for long-lived and resource-scarce systems, such as supercomputers, data centres, and device clouds that often confront energy constraints 21 – 24 . The scale of distribution is an essential element in implementing distributed computing systems due to its effectiveness. Structurally static point-to-point mechanisms suffer from intensive scaling problems because they are optimised for topology and resource capacity. The above limitations are, however, countered by virtualization that delivers a sequential and scalable platform capable of adapting in real-time and/or accommodating new and differing workloads or growing network dimensions 25 . Aggregation in edge-enriched networks, characterised by highly variable numbers of connected devices and data volume, also means that it is very easy to scale virtualization in terms of added or removed virtual resources 26 . The flexibility allows growth across the system and, at the same time, enables it to handle increased load without negatively impacting the system’s quality 27 – 31 . Security and privacy are significant challenges in distributed computing, mainly when data is processed or transmitted within an organisational environment 32 . Through virtualization, tasks and resources are run in instances that create a barrier to accessing or manipulating the system or data. For example, in multi-tenancy, virtualization guarantees that the jobs owned by different users or organisations run independently to minimise the risk of such intrusions or collisions. Also, virtualization enhances the procedures of implementing secure computing protocols like encryption and authentication to secure data and enable secure connections between nodes. Edge computing processes data close to where it came from, which makes response times faster than in standard cloud-based architectures. However, this change does come with problems, like less processing power at the edge of the network and the need for devices at the edge to work well with cloud-based services. But virtualization and other abstraction layers can add extra work that could slow down the system. Implementing resource consolidation, dynamic task management, and unified operational interfaces can all help solve these problems. When configuring virtualized environments, several critical factors must be carefully considered, including compatibility with diverse hardware and operating systems, scalability across large networks, and the overall security of the system. Virtualization is very important in healthcare systems because it makes it possible to assign tasks correctly, use resources efficiently, and quickly diagnose and treat patients. In the context of new healthcare environments, the main goal of this study is to create a model that shows how virtualization overhead affects how well tasks can be offloaded. Using SimPy, we simulate system metrics like how long it takes to finish a task, how many resources are used, how much energy is used, and how scalable the system is. The research also looks at smart systems that use edge computing, cloud services, and Internet of Things devices to help people make decisions based on data, manage patients better, and have remote consultations. Task offloading is the process of spreading computing tasks across multiple nodes. This makes the system more flexible and faster. However, the abstraction methods needed for sharing resources and keeping them safe also add to processing costs, which could affect healthcare applications that are sensitive to latency. To get the most out of the resources we have and make sure that modern healthcare systems work reliably and in real time, we need to know how these trade-offs work in numbers. Table 1 presents a comparative summary of representative task offloading techniques The paper is inspired by the significant risks that virtualization poses to flexibility and isolation in multi-kernel task offloading schemes, while simultaneously incurring the penalty of poor performance. The research endeavours to make a comparative assessment of virtualised and non-virtualised systems through a scenario-based simulation analysis, specifically in the healthcare domain. Table 1. Comparison of existing task offloading techniques. Technique Advantages Limitations Applications Static allocation 33 Simple, low computational complexity Poor adaptability to dynamic workloads General-purpose distributed systems Dynamic scheduling 34 Improved resource utilization and system efficiency Additional computational overhead Distributed computing platforms Energy-aware offloading 5 Reduced overall energy consumption Dependence on accurate energy models IoT and wireless sensor networks Containerization 35 Lightweight and scalable virtualization Potentially weaker isolation compared to hypervisor-based VMs Cloud-native and microservices-based systems Edge Computing 36 Low latency for real-time applications Limited computational capacity at edge devices Healthcare and IoT applications Cloud–edge collaboration 37 Combines high computational capacity with reduced latency Requires complex orchestration and resource management Smart cities and healthcare systems Open in a new tab Contributions The research contributes to smart healthcare domain in the following ways: Development of a simulation-based framework to incorporate virtualization overhead in task offloading scenarios for healthcare environments. A statistically rigorous evaluation of performance trade-offs introduced by virtualization that includes task completion time, energy consumption, throughput, scalability, and system reliability. Implementation and comparison of multiple scheduling strategies, that includes the proposed Virtualization-Aware Optimized Scheduling, Round-Robin, Min-Min, and HEFT. Designing and evaluating fault-tolerant mechanisms based on controlled failure injection and virtual machine migration modeling. Providing a reproducible experimental framework through controlled random seeding, full parameter disclosure, and public release of simulation code. Problem formulation The task offloading problem is formulated as a multi-objective optimization problem that minimizes task completion time and energy consumption while maximizing resource utilization and system reliability in the presence of virtualization overhead. System model Tasks and nodes Let denote the set of m computational tasks. Let denote the set of k heterogeneous processing nodes. Each task is characterized by: : Computational workload (CPU cycles). : Deadline constraint. : Energy requirement (Joules). Each node is characterized by: : Processing capability (CPU cycles per second). : Available communication bandwidth. Virtualization overhead The virtualization overhead is modelled as a linear function: where represents the fixed initialization cost and represents the proportional per-cycle overhead. The linear overhead model captures both constant startup latency and workload-dependent costs introduced by the virtual layer, consistent with empirical observations in prior virtualization performance studies. The parameters and are calibrated through micro-benchmarking and sensitivity analysis, as described in the experimental section. Objective function The objective of the optimisation is to minimise the total task completion time, , while balancing energy consumption, , and maximising resource utilisation, U 38 – 40 . Task completion time The task completion time for task offloaded to node is given by: where is the data size to be transferred. The total task completion time is: where is a binary decision variable: Energy consumption The energy consumption for task on node is: where is the idle power of the node, and is a proportional energy factor. The total energy consumption is: Resource utilization Resource utilisation for node is given by: The average utilisation across all nodes is: Constraints The problem is subject to the following constraints: Task allocation Each task must be assigned to exactly one node: Node capacity The total workload assigned to a node must not exceed its capacity: Deadline constraint Each task must be completed within its deadline: Optimization problem The optimisation problem can be formulated as: Where , , and are weight factors balancing the trade-off between completion time, energy efficiency, and resource utilisation, respectively. here, set , , and to prioritize time and energy efficiency. The optimization problem is addressed through simulation-driven techniques outlined in the methodology section. The findings illustrate the system’s capability to effectively manage the trade-offs among task completion time, energy consumption, and resource utilization. Proposed methodology To improve clarity and provide a holistic view, Fig. 1 presents a high-level architectural overview of the proposed virtualization-aware task offloading framework. The figure visually summarizes the interaction between IoT healthcare devices, virtualization-enabled edge/cloud nodes, task scheduling, and performance evaluation modules, serving as a roadmap for the detailed system modeling and algorithmic descriptions that follow. The proposed methodology shown in Fig. 1 aims to model and evaluate task-offloading mechanisms in a smart healthcare environment, explicitly considering the impact of abstraction-layer overhead on system performance. The methodology is structured into several interrelated components, including system modeling, task offloading, scheduling, performance metrics, and simulation workflow. Fig. 1. Open in a new tab Proposed methodology. System model The healthcare system is modeled as a distributed network of N computational nodes: where each node has a computational capacity and can execute tasks offloaded from IoT devices or healthcare applications. A total of T tasks, arrive randomly for processing, following a Poisson process with rate . Each task requires CPU cycles. Task offloading and virtualization overhead The base processing time for a task on node is given by: In a virtualized environment, additional overhead arises due to virtualization , modeled as: where represents fixed overhead and captures task-dependent overhead. Therefore, the total execution time in a virtualized setting becomes: Task scheduling Tasks are assigned to nodes using the Virtualization-Aware Greedy Optimization (VAGO) algorithm, which minimizes a combined objective function: where is the assignment matrix element: The task completion time accounts for computation, communication, and virtualization overhead: where is the data transfer time for task . Simulation tools and environment The simulation has been developed in Python using the SimPy framework, a discrete-event simulation library. the worksflow is shown in figure 2 To ensure statistical reliability, 30 independent runs were conducted for each scenario. The complete simulation code along with an extended set of parameters will be made publicly available upon publication to allow full reproducibility. Key parameters used in the simulation are summarized in Table 2 . Fig. 2. Open in a new tab Simulation workflow. Table 2. Simulation Parameters and Distributions. Parameter Symbol Value/distribution Description Number of nodes N 10 Available processing nodes Total tasks (per run) T 100 Total tasks generated for each run Task workload Uniform( , ) cycles Computational load (CPU cycles) Node processing power Uniform(10, 50) cycles/s Processing capacity Node bandwidth Uniform(100, 200) MB/s Communication speed Comm. data size Uniform(1, 10) MB Data size for offloading Fixed virt. overhead 0.05 s Fixed delay for VM management Prop. virt. overhead Overhead per computational cycle Fault injection rate Rate 5% to 20% (Poisson) Probability of node failure Migration factor 0.1 Proportion of work added as overhead Objective weight 1 0.4 Weight for Completion Time Objective weight 2 0.4 Weight for Energy Consumption Objective weight 3 0.2 Weight for Resource Utilization Open in a new tab The performance metrics are calculated as the mean across 30 runs for both the proposed (Virtualization-Aware Optimized) and baseline (Traditional Round-Robin) systems. Algorithm 1. Open in a new tab Virtualization-aware simulation workflow with statistical rigor. The methodology provides a structured approach to evaluate the impact of virtualization on smart healthcare task offloading systems, offering practical insights for system optimization. Algorithm 2. Open in a new tab Virtualization-aware task offloading algorithm. Proposed virtualization-aware task offloading algorithm The offloading technique proposed in the paper for task offloading in a virtualised innovative healthcare system is provided as a method. The main objective is to achieve fast task turnover while considering energy usage and optimal resource utilisation. The Virtualization-Aware Task Offloading Algorithm (Algorithm 2) improves the distributed computing paradigm by dynamically and adaptively handling the task offloading problem in virtualised environments. The algorithm takes advantage of virtualization to provide a real-time view of resources, network, and workload, which helps in performing task assignment wisely at edge and cloud points. It also differs from other traditional methods that predetermine resource distribution. Once the task is assigned, it may be re-assigned or migrated to nodes that offer optimum performance, low power consumption costs, and minimal Latency. In another aspect, the algorithm applies predictive analytics to proactively adjust the workload distribution that may be a cause of bottlenecks in the resource. In conjunction with task scheduling, the method enhances system dependability and adaptability, primarily for research-grade distributed computing applications. The application shows measurable and qualitatively significant improvements in throughput, resource utilisation, and scalability, making it a candidate to be a base technology for future distributed systems. Results and discussion Statistical validation and experimental setup To guarantee the robustness and reproducibility of our findings, all results displayed below represent the mean values derived from 30 independent simulation runs. The figures now show the mean values of the performance data points, along with 95% confidence intervals (error bars) to show how much they vary. The proposed Virtualization-Aware Optimized System (using Algorithm 2) is compared to the Traditional Round-Robin System, which is a well-known static baseline. Fault tolerance mechanism In the simulation, node failures are added as a Poisson process based on the “Failure Rate (%)” (Table 2 ) that has been set. The proposed system has a fault tolerance feature that uses live VM migration. When a node fails, all active tasks on that node are immediately paused and moved to the next best available node, . Using the optimization criteria from Algorithm 2 on the remaining task workload, we choose node . The recovery time includes a migration overhead ( ), which is based on the amount of work that is still left to be done on the task and the amount of data that needs to be transferred for migration. . The task is only considered done if it is finished on the new node before its original deadline, . Task completion time The elapsed time to complete a task shows how long it takes for a system to handle a certain amount of work. In conventional systems, work is impeded by resource pre-allocation and bottlenecks caused by the uneven distribution of workload. A virtualization-based system that spreads resources across nodes is used to solve the problem 41 . For example, when there are too many jobs to do, like in virtual machines (VMs), the resources can move the jobs to an inactive node or extend them to reduce the delay. The results in the figure 3 show that the average time it takes to finish a task goes down by mean when the workload is moderate to high. The suggested system was quicker at finishing tasks in all cases where the levels of difficulty were the same. Pre-emptive resource forecasting has made it possible to fix the available resources where they are most likely to be needed. The virtualized system showed that it would not be very affected by the worst-case distributions because it uses parallel processing. Fig. 3. Open in a new tab Mean comparison of task completion time with 95% confidence intervals. Resource utilization Resource utilization shows how well a system uses its resources, like memory, CPU, storage, and so on. In conventional terms of reference, static allocation leads to either resource wastage or excessive utilization 42 . The system is based on virtualization and regularly checks system parameters to move workloads around to different resources. Virtualization and processing optimization mean that workloads can run on fewer physical machines, which gets rid of unused capacity. The proposed system kept the mean utilization between 85% and 90% (with narrow confidence intervals) as shown in figure 4 . The other system had a mean utilization of 60% to 70%. The real-time monitoring of resource use and the dynamic distribution of loads in a virtualized system are what made IT infrastructure so much better. Fig. 4. Open in a new tab Mean resource utilization comparison with 95% confidence intervals. Fault tolerance To make sure that a distributed system is always available, it must be able to handle faults. When nodes fail, traditional systems often have service interruptions, which can lead to longer recovery times or lost data. The work uses VM migration and dynamic task reallocation to create a fault-tolerant system that works well when nodes fail. Figure 5 shows how the percentage of system reliability changes as failure rates go up. The suggested system stays reliable between 85% and 95% even when failure rates go up. The resilience is made possible by always copying important data and tasks across different nodes. Also, the virtualization layer keeps an eye on the health of the system in real time and moves workloads to healthier nodes when problems are found. Fig. 5. Open in a new tab Fault tolerance comparison with 95% confidence intervals. Latency Latency is the time it takes for a computer to do something. Because remote data processing happens, traditional centralized systems often have higher latency. The edge-based system that has been shown processes tasks closer to the data source, which cuts down on the amount of connection overhead. If the latency for task i is shown by the symbol , you can find the average latency like this: This is how to find the average: where N is the total number of tasks. The figure 6 showed that the virtualization-based framework cuts the average latency by 25–30%. This makes it perfect for real-time applications like streaming data analytics and Internet of Things apps. Fig. 6. Open in a new tab Latency comparison with 95% confidence intervals. Energy consumption Energy consumption is a critical performance metric in distributed computing systems, particularly in edge computing. Traditional architectures tend to consume more energy due to idle nodes and inefficient scheduling. The total energy usage of the system can be expressed as: where N is the number of nodes, and are the power consumed by node i in active and idle states, respectively, and and are the corresponding durations. In a virtualization-based system, resources are dynamically allocated according to the workload, allowing tasks to be consolidated on fewer nodes. The energy optimization can be formulated as: where M is the number of active virtual machines, is the power consumed by VM j , is its execution time, is the assigned workload, and is the processing capacity of the VM. The system can cut its energy use by 25% to 35% by using strategies like suspending virtual machines and putting nodes to sleep. This shows that it is both more efficient and more environmentally friendly. Latency is the term for the time it takes to do something on a computer. Traditional centralized systems tend to have more delay because they process data from a distance. The edge-based system that has been shown means that tasks are done closer to the data source, which cuts down on the amount of connection overhead. If the latency for task i is represented by the symbol , you can use the following formula to find the average latency: The equation is used to figure out how many tasks there are, which is N . The tests show that the virtualization-based framework lowers the average latency by 25–30%. Because of this, it is a great choice for real-time applications like streaming data analytics and Internet of Things-related apps. Energy use is an important factor to consider when judging distributed computing systems, especially edge computing. Because of idle nodes and poor scheduling, traditional architectures often use more energy. Let the total energy be where N is the number of nodes and P and T are the power and time in active and idle states. The suggested virtualization-based system automatically assigns resources based on workload, combining tasks on fewer nodes. Energy optimization can be written as where M is the number of active VMs, is their power, is their workload, and is their capacity. The system cuts energy use by 25% to 35% by using methods like VM suspension and node hibernation. This shows that it is more efficient and sustainable. The edge-based system that has been shown processes tasks closer to the data source, which cuts down on the amount of connection overhead. If the latency for task i is shown by the symbol , you can find the average latency like this: This is how to find the average: where N is the total number of tasks. The figure 7 showed that the virtualization-based framework cuts the average latency by 25–30%. This makes it perfect for real-time applications like streaming data analytics and Internet of Things apps. Fig. 7. Open in a new tab Energy consumption comparison with 95% confidence intervals. Throughput Throughput is the amount of work to be accomplished in a given period and is a direct measure of system performance. The throughput comparison is displayed in the Fig. 8 . The virtualised system nearly always exhibits a higher total throughput for different types of workloads while completing 40% more tasks per period compared to the traditional computing system. This improvement is attributed to the virtualisation system’s ability to dynamically acquire unused resources for its processes and efficiently use parallelism. Fig. 8. Open in a new tab Throughput comparison with 95% confidence intervals. However, if nodes are added to the system, which was designed as a modular system, the throughput continues to increase dramatically. Taken together, the Fig. 8 highlight the system’s remarkable capacity to handle large volumes of work characteristic of Big Data applications. Scalability Scalability means the system’s capacity to grow in response to the growing demands of work, or in other words, the extent to which it can be made larger. In most traditional systems, scalability is a significant problem because the architectures are usually rigid and are fixed to their associated hardware. However, the proposed virtualisation-based system is naturally scalable to adapt to the increase in nodes or more VMs. From the Fig. 9 , even as the workloads increased to double and triple the current level, the proposed system recorded 85% efficiency. On the other hand, traditional systems have numerous constraints and suffer a drastic drop in efficiency due to bottlenecks. The virtualisation layer allows for other resources to be allocated differently so that the new workloads are balanced across the network. This makes the system versatile when facing other conditions, including increased traffic loads or the incorporation of different applications. Fig. 9. Open in a new tab Scalability comparison: mean utilization efficiency vs. workload increase. Sensitivity analysis of virtualization overhead ( ) A sensitivity analysis was performed by varying the fixed virtualization overhead ( ) from 0.01 to 0.10 to understand its impact on the mean Task Completion Time ( ). The results showed that for low workloads (10 tasks), the completion time increased by up to when was at its maximum (0.10 s), compared to the minimum (0.01 s). However, for high workloads (40 tasks), the impact was mitigated, with only a increase in . It confirms the optimization algorithm’s ability to efficiently amortize the fixed overhead cost across parallel tasks, maintaining high performance even in environments with non-trivial virtualization costs. Conclusion This study investigated virtualization-aware task offloading in smart healthcare environments using a simulation-driven framework. The proposed system was evaluated against a baseline approach across multiple performance indicators, including average task completion time, energy efficiency, fault tolerance, scalability, resource utilization, and virtualization overhead. The results demonstrate that incorporating virtualization-aware optimization significantly improves system reliability, utilization efficiency, and energy performance, making the approach suitable for dynamic and resource-intensive healthcare applications. At the same time, the analysis highlights the non-negligible overhead introduced by virtualization layers, particularly in terms of task execution latency, underscoring the need for careful overhead-aware system design. Future research will focus on extending the proposed framework by incorporating real-world healthcare workloads and heterogeneous virtualization technologies such as containers and lightweight microVMs to further reduce overhead. Additionally, adaptive learning-based schedulers can be explored to dynamically tune virtualization parameters in response to workload variations and failure patterns. Investigating security- and privacy-aware offloading strategies and validating the approach on real healthcare edge–cloud testbeds represent important directions to address the current simulation-based limitations and enhance practical deployment feasibility. Acknowledgments This research work is funded by Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2026R40), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia Author contributions Gaurav Dhiman wrote the paper, Kiran Deep Singh simulated the work, Prabh Deep Singh analyzed the work, Norah Saleh Alghamdi simulated the work, Ghadah Shukri Albakri supervized the work. Funding Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2026R40), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia Data availability No datasets were generated or analysed during the current study. The simulation code and synthesized datasets used to generate the figures will be made publicly available via a persistent repository (e.g., GitHub and Zenodo) upon acceptance of the paper for full reproducibility. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References 1. Hartmann, M., Hashmi, U. S. & Imran, A. Edge computing in smart health care systems: Review, challenges, and research directions. Trans. Emerg. Telecommun. Technol. 33 (3), e3710 (2022). [ Google Scholar ] 2. Aminizadeh, S., Heidari, A., Toumaj, S., Darbandi, M., Navimipour, N. J., Rezaei, M., Talebi, S., Azad, P. & Unal, M. The applications of machine learning techniques in medical data processing based on distributed computing and the internet of things. In Computer Methods and Programs in Biomedicine . 107745 (2023). [ DOI ] [ PubMed ] 3. Harjula, E., Kumar, T., Islam, J., Ejaz, M. & Kovacevic, I. Distributed network and service architecture for future digital healthcare. Finnish J. eHealth eWelfare 14 (1), 6–18 (2022). [ Google Scholar ] 4. Singh, K., Singh, P. & Angurala, M. Future directions in healthcare research. In Computational Health Informatics for Biomedical Applications . 287–313. (Apple Academic Press, 2023). 5. Mansouri, Y. & Babar, M. A. A review of edge computing: Features and resource virtualization. J. Parallel Distrib. Comput. 150 , 155–183 (2021). [ Google Scholar ] 6. Ranjit, K. C., Shrestha, B. & Ghimire, U. Optimizing cnn architectures for steering angle prediction for self-driving vehicles in unstructured roads: A comparative study of activation functions and model complexity. ICCK Trans. Mach. Intell. 2 (2), 88–99 (2026). [ Google Scholar ] 7. Kumar, A. & Lehal, G. S. Detection of newspaper layouts using yolo12. ICCK Trans. Mach. Intell. 2 (2), 77–87 (2026). [ Google Scholar ] 8. Wang, X. & Zhao, R. Fedtlrec: Federated recommendation with transformer-based parameter aggregation and Lora compression. ICCK Trans. Mach. Intell. 2 (1), 65–76 (2026). [ Google Scholar ] 9. Arya, S., Anju, & Onah, J. N. A data-driven framework for methane emission prediction using machine learning methods. In ICCK Transactions on Machine Intelligence . Vol. 2(1). 53–64 (2026). 10. Bai, Y., Hu, J., Tursun, E. & Yimit, H. Data-driven operational assessment method and digital twin system for unmanned surface vehicles. ICCK Trans. Mach. Intell. 2 (1), 38–52 (2026). [ Google Scholar ] 11. Lakshminarayanan, V., Ravikumar, A., Sriraman, H., Alla, S. & Chattu, V. K. Health care equity through intelligent edge computing and augmented reality/virtual reality: A systematic review. J. Multidiscip. Healthc. 2839–2859 (2023). [ DOI ] [ PMC free article ] [ PubMed ] 12. Agarwal, S. K. & Gupta, Y. K. A novel approach of progressive transfer learning for MRI brain tumor classification using vgg16 and MobileNET architectures. ICCK Trans. Mach. Intell. 2 (1), 28–37 (2026). [ Google Scholar ] 13. Kumar, V., Saxena, A. & Chawla, L. Adaptive learning density estimators for Tsallis entropy and Kapur entropy with applications in system training. ICCK Trans. Mach. Intell. 2 (1), 12–27 (2026). [ Google Scholar ] 14. Agarwal, S., Kandpal, M., Gupta, V. K., Aggarwal, S. & Jain, P. Secure and decentralized heart sound analysis using federated learning and blockchain technology. ICCK Trans. Mach. Intell. 2 (1), 1–11 (2025). [ Google Scholar ] 15. Pereira, M. S. et al. Discriminating planted Capsicum spp. varieties via machine learning and multivariate data reduction. ICCK Trans. Mach. Intell. 1 (3), 166–185 (2025). [ Google Scholar ] 16. Surendran, R. & Tamilvizhi, T. Cloud of medical things (COMT) based smart healthcare framework for resource allocation. (2021). 17. Kumar, A. Clinical text analytics: Techniques, deep learning models, and the future of medical text analytics. ICCK Trans. Mach. Intell. 1 (3), 148–165 (2025). [ Google Scholar ] 18. Malik, I. A. & Habib, G. Ai enabled resource-constrained computing architectures for IOT devices. ICCK Trans. Mach. Intell. 1 (3), 138–147 (2025). [ Google Scholar ] 19. Bmotra, M., Kumar, R. & Dhiman, G. Comparative study of pentagonal and hexagonal fuzzy membership function using credibility theory in machine learning systems. ICCK Trans. Mach. Intell. 1 (3), 127–137 (2025). [ Google Scholar ] 20. Aggarwal, V., Aggarwal, S., Bhardwaj, A. & Gupta, V. K. Quantum computing essentials: Bridging theory and practice for new learners. ICCK Trans. Mach. Intell. 1 (3), 117–126 (2025). [ Google Scholar ] 21. Sarvakar, K. & Rana, K. A hybrid framework combining cnn, lstm, and transfer learning for emotion recognition. ICCK Trans. Mach. Intell. 1 (2), 103–116 (2025). [ Google Scholar ] 22. Singh, S., Aggarwal, S., Singh, A. & Sharma, A. Integrating artificial intelligence and machine learning in autism detection via gut microbiome analysis. ICCK Trans. Mach. Intell. 1 (2), 90–102 (2025). [ Google Scholar ] 23. Sharma, S., Devgan, K., Jangra, D., Bhardwaj, A. & Aggarwal, S. Neuro-inspired alert system for air quality prediction using ensemble preprocessing and snn classification. ICCK Trans. Mach. Intell. 1 (2), 69–79 (2025). [ Google Scholar ] 24. Khan, A. & Singh, J. A novel image captioning technique using deep learning methodology. ICCK Trans. Mach. Intell. 1 (2), 52–68 (2025). [ Google Scholar ] 25. Vyas, S., Bhargava, D. & Khan, S. Healthcare 4.0: A systematic review and its impact over conventional healthcare system. In Artificial Intelligence for Health 4.0: Challenges and Applications . 1–17 (2023). 26. Javeed, D., Gao, T., Saeed, M. S., Kumar, P., Kumar, R. & Jolfaei, A. A softwarized intrusion detection system for IOT-enabled smart healthcare system. In ACM Transactions on Internet Technology (2023). 27. Khan, M., Kumar, R. & Dhiman, G. A hybrid machine learning fuzzy non-linear regression approach for neutrosophic fuzzy set. ICCK Trans. Mach. Intell. 1 (1), 42–51 (2025). [ Google Scholar ] 28. Garg, M. Enhanced reinforcement learning-based resource scheduling for secure blockchain networks in IIOT. ICCK Trans. Mach. Intell. 1 (1), 29–41 (2025). [ Google Scholar ] 29. Gupta, S., Adhikari, U., Roy, D. & Hazra, S. IOT-integrated reinforcement learning-based mine detection system for military and humanitarian applications. ICCK Trans. Mach. Intell. 1 (1), 17–28 (2025). [ Google Scholar ] 30. Myakala, P. K., Kamatala, S. & Bura, C. Privacy-preserving federated learning for IOT botnet detection: A federated averaging approach. ICCK Trans. Mach. Intell. 1 (1), 6–16 (2025). [ Google Scholar ] 31. Tiwari, S., Kumar, D., Mahajan, A. & Sachar, S. Emotion detection from speech using cnn-bilstm with feature rich audio inputs. ICCK Trans. Mach. Intell. 1 (2), 80–89 (2025). [ Google Scholar ] 32. Singh, P. D., Singh, K. D., Christa, S. & Dangi, S. Redefining healthcare with quantum computing. In Recent Advances in Computing Sciences (2025). 33. Yang, S., Li, F., Trajanovski, S., Yahyapour, R. & Fu, X. Recent advances of resource allocation in network function virtualization. IEEE Trans. Parallel Distrib. Syst. 32 (2), 295–314 (2020). [ Google Scholar ] 34. Hu, F., Mehta, K., Mishra, S. & AlMutawa, M. A dynamic distributed scheduler for computing on the edge. In 2024 International Symposium on Parallel Computing and Distributed Systems (PCDS) . 1–7 (IEEE, 2024). 35. Bentaleb, O., Belloum, A. S., Sebaa, A. & El-Maouhab, A. Containerization technologies: Taxonomies, applications and challenges. J. Supercomput. 78 (1), 1144–1181 (2022). [ Google Scholar ] 36. Alves, M. P., Delicato, F. C., Santos, I. L. & Pires, P. F. Lw-coedge: A lightweight virtualization model and collaboration process for edge computing. World Wide Web 23 (2), 1127–1175 (2020). [ Google Scholar ] 37. Zangana, H. M., khalid Mohammed, A., & Zeebaree, S. R. Systematic review of decentralized and collaborative computing models in cloud architectures for distributed edge computing. Sist. J. Sist. Inf. 13 (4), 1501–1509 (2024). 38. Ghafari, R., Kabutarkhani, F. H. & Mansouri, N. Task scheduling algorithms for energy optimization in cloud environment: A comprehensive review. Cluster Comput. 25 (2), 1035–1093 (2022). [ Google Scholar ] 39. Kruekaew, B. & Kimpan, W. Multi-objective task scheduling optimization for load balancing in cloud computing environment using hybrid artificial bee colony algorithm with reinforcement learning. IEEE Access 10 , 17803–17818 (2022). [ Google Scholar ] 40. Hu, B., Cao, Z. & Zhou, M. Energy-minimized scheduling of real-time parallel workflows on heterogeneous distributed computing systems. IEEE Trans. Serv. Comput. 15 (5), 2766–2779 (2021). [ Google Scholar ] 41. Mutlag, A. A., Khanapi Abd Ghani, M., Mohammed, M. A., Maashi, M. S., Mohd, O., Mostafa, S. A., Abdulkareem, K. H., Marques, G. & de la Torre Díez, I. Mafc: Multi-agent fog computing model for healthcare critical tasks management. Sensors 20 (7), 1853 (2020). [ DOI ] [ PMC free article ] [ PubMed ] 42. Velez, F. F., Colman, S., Kauffman, L., Ruetsch, C. & Anastassopoulos, K. Real-world reduction in healthcare resource utilization following treatment of opioid use disorder with reset-o, a novel prescription digital therapeutic. Expert Rev. Pharmacoecon. Outcomes Res. 21 (1), 69–76 (2021). [ DOI ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Data Availability Statement No datasets were generated or analysed during the current study. The simulation code and synthesized datasets used to generate the figures will be made publicly available via a persistent repository (e.g., GitHub and Zenodo) upon acceptance of the paper for full reproducibility. Articles from Scientific Reports are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (2.7 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top