A scalable scheduling and resource management framework for cloud-native B2B applications Authors: Dinesh Komarasamy, Rajkumar Rajavel, Karthikeyan Harimoorthy, Aravind Pitchai Published: 2025 ABSTRACT In modern cloud computing environments, customers increasingly depend on on-demand resource provisioning to handle dynamic workloads. However, fluctuations in job arrival rates can result in prolonged queue times, which negatively affect overall system performance. Although existing scheduling algorithms provide efficient job management, they often fail to account for the combined impact of queue delays and the need for flexible resource provisioning—particularly in business-critical applications. In order to tackle these issues, the paper proposes a new Optimized Job Scheduling and Resource Scaling (OJSRS) algorithm designed to improve job execution efficiency and support elastic resource management in cloud environments. The OJSRS algorithm integrates two key components: Tree-based Job Scheduling (TJS) and Automated Resource Scaling and Scheduling (ARSS). The TJS component constructs a hierarchical structure that concurrently maps incoming jobs to the most suitable Virtual Machines (VMs), thereby minimizing queue delays. Meanwhile, ARSS adjusts resource allocation dynamically, increasing or decreasing capacity according to workload requirements and cloud service provider policies, enabling responsive and adaptive provisioning. Experimental results show that the OJSRS algorithm increases resource utilization by approximately 5–10% and accelerates job completion through proactive resource scaling. This approach provides a significant performance advantage for cloud-native business applications that require both efficiency and scalability. Introduction Cloud computing brings together a wide range of distributed resources to handle both compute-intensive and data-intensive tasks1. Business applications are also among the compute-intensive tasks executed in a cloud environment. The cloud service model is commonly divided into three layers: IaaS, PaaS, and SaaS. Among these, proposed study focuses on IaaS, which provides extensive computing and storage resources. In contrast, PaaS provides a development and deployment environment tailored to user needs, whereas SaaS delivers ready-to-use software applications to end users2. Most novice users prefer cloud environments for running jobs because the cloud enables on-demand resource provisioning, resource resizing, job migration, quick elasticity, extensive network contact, resource sharing, metered services, and improved availability and performance3. However, cloud environments still face challenges in interoperability, security, and the absence of built-in automated schedulers. Cloud infrastructure leverages virtualization by partitioning computing capacity into multiple logical units, i.e., Virtual Machines (VMs). Virtualization technology allows dynamic VM creation, resizing, migration, and termination at runtime4. Many Cloud Service Providers (CSPs) deliver computational power to users for processing their jobs. However, traditional CSPs have a limited pool of resources, which leads to overprovisioning when job submissions are low and under provisioning when job submissions are high5. Figure 1 illustrates the agreement on service levels established between CSPs and end users. Individual users or groups of users connect to the cloud to enhance their business through online platforms. The SLA specifies terms such as usage policies, processing speed, and cost. Thus, scalability of computational resources is critical to meeting user expectations and SLA requirements. An SLA represents a time and cost agreement between the user and the service provider6. CSPs employ both elastic and non-elastic provisioning strategies to fulfill SLA requirements7,8. Elastic provisioning executes jobs by creating and deleting VMs at runtime for each job, which may lead to excessive memory usage. In contrast, non-elastic provisioning executes jobs using only the VMs in the resource pool, often causing underutilization or overutilization. Fig. 1Service Level Agreement. Service Level Agreement. Effective scheduling algorithms are essential to balance workloads across VMs and ensure timely job completion. However, job and resource scheduling remains challenging, as CSPs aim to predict the economic use of computational resources while end users demand high-quality, cost-effective services9,10. Existing scheduling algorithms typically process jobs sequentially and often fail to account for queue delays. Furthermore, traditional CSPs usually support either elastic or non-elastic provisioning, which can negatively affect Quality of Service (QoS) or resource utilization. Traditional CSPs either support autoscaling or efficient resource utilization, but rarely both. Therefore, the goal of this research is to integrate optimal scheduling of jobs with autoscaling mechanisms. To this end, we propose a two-phase Optimized Job Scheduling and Automated Resource Scaling (OJSRS) algorithm, which consists of Tree-based Job Scheduling (TJS) and Automated Resource Scaling and Scheduling (ARSS). In the first phase, TJS concurrently maps jobs to VMs using a right-skewed binary tree constructed from currently available VMs. When the available VMs cannot meet job deadlines, the second phase scales up VMs to ensure timely execution and minimize job queuing delays. The novelty of OJSRS lies in three aspects. First, the concurrent mapping of jobs to VMs using the right-skewed tree in TJS minimizes queue delays, whereas existing approaches (e.g., EDF, MVF, AMSS, PCP) rely on sequential mapping. Second, OJSRS integrates autoscaling decisions directly with unmapped jobs identified by TJS, ensuring demand-driven scaling at the job level rather than depending solely on static utilization thresholds. Third, OJSRS incorporates two-tier VM deployment (foreground and background), enabling conditional preemption without job interruption—an approach not found in earlier hybrid methods. Fourth, job execution and resource allocation are carried out independently by deploying two-tier VMs. Tasks are executed in the foreground part, while the background part utilizes the residual idle processing speediness to prioritize and map jobs to the resources of the cloud environment. The rest of this paper is organized as follows: The section "Literature review" reviews the related literature. The section "Characteristics of system model" outlines the proposed system, including the job model, system model, and problem definition. The section "Optimized job scheduling and automated resource scaling algorithm in cloud" explains the system architecture and its components. The section "Experimental setup and result analysis" discusses the experimental results and their analysis. Finally, the section "Conclusion and future work" provides the conclusion and highlights potential future research directions. Literature review A scheduler is responsible for selecting jobs from a queue and allocating them to VMs to ensure fair and timely completion. Most schedulers employ two mechanisms: job scheduling and resource scheduling. Job and resource scheduling are critical both for meeting the needs of novice users and for fulfilling the objectives of Cloud Service Providers (CSPs). Job scheduling algorithms rearrange the execution order of jobs to maximize the number completed before their deadlines, while resource scheduling assigns jobs to the most suitable VMs in the resource pool to achieve CSP objectives11. To support novice users, researchers have developed a variety of job scheduling algorithms based on parameters such as deadline, timestamp, cost, makespan, and throughput. One of the earliest approaches was the Earliest Deadline First (EDF) algorithm, which prioritizes jobs with the earliest deadlines to maximize timely completion12. However, the EDF algorithm does not preempt a running job even if a newly arrived job has an earlier deadline. To address this, the fully preemptive EDF (fp-EDF) procedure was presented, allowing preemption of currently running jobs. This was further refined into the controlled preemptive EDF (cp-EDF) procedure, that conditionally preempts jobs only when necessary, thereby improving resource utilization by avoiding unnecessary interruptions13. Despite these improvements, both fp-EDF and cp-EDF may lead to starvation, as jobs with the earliest deadlines dominate execution. To mitigate this, the Minimum Variation First (MVF) algorithm was developed, which schedules jobs based on the minimum deviation calculated by subtracting the deadline from the expected execution time. Expected execution time is derived from job attributes such as arrival time, length, and deadline14. Another approach, the Adaptive Multilevel Scheduling System (AMSS), schedules jobs by considering both job parameters and VM processing speed, sequentially assigning jobs to the most suitable VMs in the pool15. However, this approach may still fail to complete some jobs within their deadlines. To address this, the Partial Critical Path (PCP) algorithm was proposed. PCP operates in two phases: distribution, where deadlines are distributed to identify the critical path, and planning, where sub-deadlines are assigned and jobs are mapped to the least costly resources. The PCP algorithm was later extended into the IaaS Cloud PCP (IC-PCP), a single-phase approach that schedules jobs directly rather than assigning sub-deadlines16. To further reduce deadline violations, Enhanced Infrastructure Cloud PCP with Replication (EIPR) was introduced, which executes jobs redundantly across multiple VMs. Although effective in preventing deadline misses, this often results in resource underutilization17. Several resource scheduling algorithms have also been developed to improve job-to-VM mapping. For example, the Dominant Resource Fairness (DRF) method clusters and schedules jobs based on dominant resource factors such as storage, computing power, or network capacity18. The Resource Attribute Selection (RAS) algorithm, instead, jointly considers computing capacity, storage, and network bandwidth to allocate resources efficiently. RAS also incorporates a work-stealing technique to balance workloads among VMs19. Heuristic approaches have also been explored, including mixed integer linear programming formulations combined with best-fit and tabu search strategies to optimally map jobs to VMs20. However, such approaches typically depend on prior knowledge of job characteristics, which is often unavailable. To overcome this limitation, algorithms such as the Blind Scheduling Algorithm (BSA) and Blind Online Scheduling Algorithm (BOSA) were developed, which operate without prior job knowledge. BOSA, for instance, schedules jobs using the First Come, First Serve (FCFS) approach and assigns them to clusters with large idle time slots to minimize waiting time21,22. Additionally, a hierarchical load-balancing algorithm was proposed to distribute workloads across clusters by forwarding jobs to clusters with the highest available computing capacity. This approach considers CPU speed, memory usage, and network utilization23. Non-elastic resource provisioning is well-suited for batches of jobs with similar workloads. However, depending on job characteristics, VMs may be underutilized (leading to idle resources) or overutilized (reducing performance). Elastic resource provisioning, in contrast, is more suitable when job arrival rates vary and jobs must meet strict deadlines. In such cases, VMs can be dynamically acquired or released at runtime, ensuring better alignment with workload fluctuations and avoiding both underutilization and overutilization. Therefore, optimal resource scaling is critical in elastic provisioning scenarios24. To optimize resource use and minimize costs, several approaches leverage dynamic VM scaling. For example, policies such as the energy-aware rolling horizon scheduling algorithm dynamically create or terminate VMs in data centers to reduce energy consumption while meeting job deadlines27. Similarly, class-constrained bin packing has been applied to enhance user satisfaction through efficient resource scaling28. However, challenges remain: insufficient resources or delays in VM creation can increase waiting time, degrade performance, and add network overhead. The proposed OJSRS algorithm addresses these challenges by combining optimal job scheduling with automated resource scaling. It maximizes resource utilization through a two-tier VM deployment strategy, reduces job waiting times, and optimizes energy consumption while maintaining service quality. Characteristics of system model Job model Many users have adopted cloud computing to process jobs cost-effectively, owing to the advancement of the pay-per-usage model and on-demand resource provisioning. Since the number of jobs submitted to cloud systems varies over time, the job arrival rate is difficult to predict; therefore, incoming jobs are considered aperiodic. Moreover, detailed information about a job (i.e., its attributes) is not known in advance within the cloud system; hence, such jobs are also referred to as ‘Apriori’. Typically, newly arriving jobs are placed in a queue \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left({J}_{q}\right)$$\end{document} and processed as batch job. A batch contains a set of jobs \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{J}_{q}=\left\{{j}_{1},{j}_{2},...,{j}_{n}\right\}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:n$$\end{document} denotes the count of incoming jobs at a given time. These jobs are assumed to be independent, meaning the execution of one job \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{J}_{q}$$\end{document} in does not depend on the execution of another. During submission, the user provides two primary parameters for each job\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{i}\left(ls,d\right)$$\end{document}: the job length \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:ls\:$$\end{document}and the job deadline \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:d$$\end{document}. Each job requests a single VM for execution. Once assigned, the job is processed exclusively on that VM and is non-pre-emptive, i.e., it cannot be interrupted or migrated during execution. System model Cloud computing provides services to end users through multiple Cloud Service Providers (CSPs) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:Cc=\left\{{CSP}_{1},{CSP}_{2},...,{CSP}_{m}\right\}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:m$$\end{document} denotes the total number of CSPs. This work focuses on how a CSP can optimally serve end users. Each CSP hosts several Virtual Machines (VMs) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{CSP}_{i}=\left\{{VM}_{1},{VM}_{2},\dots\:,{VM}_{\alpha\:}\right\}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document} represents the overall count of VMs in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:CS{P}_{i}$$\end{document}, and these VMs operate independently of one another. Each VM in a CSP is characterized by two mandatory attributes: processing speed and storage capacity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left({VM}_{i}\left(ps,s\right)\right)$$\end{document}. The processing capacity of a VM varies depending on the policies of the CSP, which may support either homogeneous or heterogeneous environments. In a homogeneous environment, all VMs within a CSP have the same processing speed, i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{VM}_{i}\left(ps\right)={VM}_{j}\left(ps\right)={VM}_{k}\left(ps\right)$$\end{document}. In contrast, in a heterogeneous environment, VMs within a CSP have different processing speeds, i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{VM}_{i}\left(ps\right)\ne\:{VM}_{j}\left(ps\right)\ne\:{VM}_{k}\left(ps\right)$$\end{document}. Problem statement As represented in the system model, user requests are treated as computational tasks submitted by users. Numerous jobs request services from the cloud system, and the number of jobs varies over time. Consequently, forecasting job arrival rates is a difficult task, making job scheduling both important and complex in cloud computing. After submission, jobs are executed on the available virtual machines. In general, the cloud system supports two types of resource provisioning. In the first category, tasks are processed through the VMs provided by the resource pool. However, when the number of jobs is small, VMs are underutilized, and when the number of jobs is large, the available VMs may not provide sufficient processing capacity. In such cases, jobs may miss their deadlines, negatively impacting system performance, Quality of Service (QoS), and user satisfaction. In the second category, VMs can be created at runtime based on user requirements. After the job completes, the VM is destroyed. While this approach ensures flexibility, frequent VM creation and destruction affect resource utilization. Additionally, when a high-priority job arrives at a specific VM, the currently running job may be preempted to meet the new deadline. Although this ensures timely execution, frequent preemptions also reduce resource efficiency. To address these challenges, VMs are organized in a two-tier setup consisting of foreground and background VMs, enabling a single VM to process two jobs concurrently. However, existing algorithms require additional time to prioritize and select the best VM from the resource pool. Therefore, a new algorithm is needed to process jobs more effectively by integrating VM utilization from the resource pool with dynamic VM scaling. For this reason, we propose the Optimized Job Scheduling and Resource Scaling (OJSRS) algorithm for cloud computing. The proposed algorithm minimizes overall processing time by executing jobs concurrently, thereby improving customer satisfaction and increasing CSP profits. Optimized job scheduling and automated resource scaling algorithm in cloud In a cloud environment, jobs demand computational resources to be executed. The environment comprises multiple CSPs that provide services to clients over the Internet through web interfaces. Each CSP manages several data centers containing multiple VMs for job execution. Depending on data center policy, VMs may operate in either a homogeneous or heterogeneous environment. At any point in time, a huge count of jobs may arrive in cloud management system. Traditionally, jobs are processed either by selecting best VM from the resource pool by creating a new VM at runtime based on user requirements. However, both methods have limitations when applied independently. To overcome these drawbacks, this paper proposes the OJSRS algorithm, which integrates job scheduling with automated resource scaling. By combining these techniques, OJSRS ensures efficient resource utilization, reduced queuing delays, and improved execution performance. Fig. 2Proposed OJSRS Algorithm. Proposed OJSRS Algorithm. Figure 2 illustrates the design of the proposed OJSRS algorithm. Client requests are treated as jobs and stored in a queue, denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{bj}$$\end{document}. Jobs placed in this queue are handled by the OJSRS algorithm within the cloud environment. Initially, jobs are scheduled using the Tree-based Job Scheduling (TJS) component, and information about the available VMs is stored in the Virtual Machine Monitor (VMM). After scheduling, some jobs may remain unmapped to resources because the VMs in CSPs are unable to complete them within their deadlines. In such cases, the jobs are processed using the dynamic resource pool. In the proposed model, the VMM maintains VM information in two separate queues: (i) information about existing VMs is stored in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{f-VMM}$$\end{document}, and (ii) information about newly created VMs is maintained in the dynamically available resource pool\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{s-VMM}$$\end{document}. The least processing speed necessary to execute a job is computed based on its attributes—namely, job length, job deadline, and job arrival time. This minimum processing speed is denoted as and is expressed by Eq. (1).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{x}\left({S}_{er}\right)=\frac{{j}_{x}\left(ls\right)}{{j}_{x}\left(d\right)}\hspace{1em};\forall\:jin{Q}_{bj}$$\end{document} Here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\prime\:}x{\prime\:}$$\end{document} represents any job in the job queue. Based on the minimum job arrival rate, a fixed number of VMs is created according to the data center’s policies and stored in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{f-VMM}$$\end{document}. Using the existing VMs in the CSP, the least available processing speed of the CSP is denoted \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:CSP{\left(ps\right)}_{min}$$\end{document} and calculated by Eq. (2).2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{CSP}_{min}\left(ps\right)=\sum\:{j}_{x}\left({S}_{er}\right)\:;\forall\:j\:in{Q}_{bj}\bigvee\:\sum\:{VM}_{y}\left(ps\right);\:\forall\:VM\:in{Q}_{f-VMM}$$\end{document} The current computing capacity of the CSP is denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:CS{P}_{current}\left(ps\right)$$\end{document} and is computed using Eq. (3). Jobs are optimally mapped to VMs by constructing a tree with the TJS technique. However, after tree construction \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{f-VMM}$$\end{document}, some jobs may remain unallocated because the available VMs in are insufficient. Such jobs are executed using the VMs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{s-VMM}$$\end{document}. In this case, since the TJS technique cannot allocate a VM, a new VM is created at runtime using the AS technique, and its information is stored in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{s-VMM}$$\end{document}. The TJS technique is described in detail in the following section. 3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{CSP}_{current}\left(ps\right)=\sum\:{VM}_{y}\left(ps\right)\hspace{1em};\forall\:VMin{Q}_{VMM}$$\end{document} Tree-based job scheduler The TJS receives a batch of jobs from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{bq}$$\end{document} that change in relation to time. Jobs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{bq}$$\end{document}are processed in a FCFS manner. A VM’s idle processing rate is expressed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{S}_{idle}$$\end{document}. Initially, the TJS technique maps each job to the most suitable VM by comparing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{S}_{idle}$$\end{document}with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{S}_{er}$$\end{document}. It then constructs a right-skewed binary tree \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left(RSBT\right)$$\end{document} using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:VM\left({S}_{idle}\right)$$\end{document} in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{f-VMM}$$\end{document} of the VMM. Similarly, the TJS technique creates another right-skewed binary tree \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left(RSB{T}^{{\prime\:}}\right)$$\end{document} using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:VM\left({S}_{idle}\right)$$\end{document} in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{s-VMM}$$\end{document}. In this process, the VMs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:RSBT$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:RSB{T}^{{\prime\:}}$$\end{document}are considered parent nodes in the TJS technique. After constructing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:RSBT$$\end{document}and\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:RSB{T}^{{\prime\:}}$$\end{document}, the jobs are mapped to the respective VMs by comparing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:j\left({S}_{r}\right)$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:VM\left({S}_{idle}\right)$$\end{document}, as defined in Eq. (4). Here, jobs are mapped to the optimal VMs that can complete them within their deadlines. The mapping of jobs to VMs is denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:M\left({j}_{x},V{M}_{y}\right)$$\end{document} and expressed as follows:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:M\left({j}_{x},{VMM}_{y}\right)=\left\{\begin{array}{c}1\hspace{1em};if{j}_{x}\left({S}_{er}\right)\le\:{VM}_{y}\left({S}_{idle}\right)\\\:0\hspace{1em};otherwise\end{array}\right.\hspace{1em}\forall\:jin\hspace{0.33em}{Q}_{bq}\&\forall\:VM$$\end{document} Here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{x}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:V{M}_{y}$$\end{document} denote the corresponding job and VM, respectively. After mapping, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:V{M}_{y}\left({S}_{idle}\right)$$\end{document} is updated based on\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{x}\left({S}_{er}\right)$$\end{document}. The value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:V{M}_{y}\left({S}_{idle}\right)$$\end{document} is computed as shown in Eq. (5).5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{VM}_{y}\left({S}_{idle}\right)=\left\{\begin{array}{c}{VM}_{y}\left({S}_{idle}\right)-{j}_{x}\left({S}_{er}\right)\hspace{1em};ifM\left({j}_{x},{VM}_{y}\right)\\\:{VM}_{y}\left({S}_{idle}\right)\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em};otherwise\end{array}\right.\hspace{1em}$$\end{document} Once \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{VM}_{y}\left({S}_{idle}\right)$$\end{document} is computed, the job is scheduled onto the most appropriate VM from the resource pool using the TJS algorithm. The scheduled time of the jobs, denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{t}_{s}$$\end{document}, is stored prior to tree construction. The TJS algorithm explains tree construction using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{x}\left({S}_{er}\right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{VM}_{y}\left({S}_{idle}\right)$$\end{document}, and it also specifies how jobs are mapped to VMs. The mapping of jobs to VMs is clearly demonstrated using the skewed tree, as expressed in Fig. 3. AlgorithmTJS Algorithm. TJS Algorithm. Figure 3 illustrates the construction of a tree using the TJS technique and explains how jobs are shaped and mapped to VMs. Jobs are mapped with VMs based on the following criteria. The VMs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:RSBT$$\end{document} are considered parent nodes, and jobs are inserted as either right or left children by comparing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{S}_{er}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:VM\left(ps\right)$$\end{document}. A new job is inserted to the left sub-tree if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{S}_{er}$$\end{document} is less than or equal to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:VM\left(ps\right)$$\end{document}; otherwise, it is compared with the execution speed of VM at the next tree level. The tree \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:RSBT$$\end{document} is constructed using several VMs \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left(\alpha\:\right)$$\end{document} from the resource pool, and it contains \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:-1$$\end{document} levels. Occasionally, some jobs are not mapped to any VMs in\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:RSBT$$\end{document}; these are directed to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:RSBT{\prime\:}$$\end{document}. Each level of the tree has one parent and one or more leaves. After the tree is created using the TJS technique, jobs are assigned to their respective VMs. Each VM maintains a queue \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left({Q}_{VM}\right)$$\end{document} to store its assigned jobs. At each level, a VM processes all corresponding left child jobs. However, VMs at the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document} -level of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:RSBT$$\end{document} tree may be unable to process all jobs, which are then executed using VMs in the dynamic resource pool\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left({Q}_{s-VMM}\right)$$\end{document}. Similarly, VMs at the level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\alpha\:}^{{\prime\:}}-1$$\end{document} may also fail to process all right-child jobs when the total number of VMs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{s-VMM}$$\end{document} exceeds. In such cases, VMs must scale up using auto-scaling to process the right-child of the node at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\alpha\:}^{{\prime\:}}-1$$\end{document} level. The rationale behind this structure is that the right-skewed binary tree (RSBT) ensures jobs requiring higher processing speeds are systematically compared down the tree, which is particularly efficient for heterogeneous workloads. The RSBT structure reduces average search and mapping overhead because jobs that cannot be handled by lower-level VMs (with lower idle speeds) are quickly escalated to higher-capacity VMs along the right branch. Alternative structures, such as balanced binary trees or heaps, would require additional restructuring operations during dynamic job arrivals, resulting in higher scheduling latency. In contrast, the RSBT maintains simplicity in insertion and job-to-VM comparison while ensuring that jobs with tighter deadlines and higher computational requirements are prioritized. Fig. 3Tree Construction using TJS Technique. Tree Construction using TJS Technique. Automated resource scaling and scheduling Due to the unpredicted arrival of jobs, VMs must scale up or down dynamically to minimize the overall expenditure of the CSP. Jobs assigned to the right side of the node at level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\alpha\:}^{{\prime\:}}-1$$\end{document} cannot be processed using VMs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{f-VMM}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{s-VMM}$$\end{document}. These unmapped jobs are placed in a queue \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{nj}$$\end{document}. In such cases, the CSP must scale up VMs to process and complete the within their deadlines. A processing speed of a scaled VM depends on data center’s policies in a homogeneous environment. In contrast, in a heterogeneous environment, VMs are created according to minimum required execution speed of the job. The minimal extra execution speed required for unmapped jobs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{nj}$$\end{document} is denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{A}_{uj}$$\end{document}, which is determined using various job attributes and computed as shown in Eq. (6).6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{A}_{uj}\left(ps\right)=\sum\:j\left({S}_{er}\right)\hspace{1em};\forall\:jin{Q}_{nj}$$\end{document} The total processing speed of scaled-up VMs is denoted as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{A}_{uj}$$\end{document}. In a homogeneous setting, the count of VMs deployed depends on the data center’s policies\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left(i.e.\eta\:=\frac{{A}_{uj}\left(ps\right)}{VM\left(ps\right)}\right)$$\end{document}. In contrast, in a diverse setting, the count of VMs deployed depends on the requirements \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{S}_{er}$$\end{document} of the jobs. Once VMs are provisioned, their information is updated in\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{s-VMM}$$\end{document}, and the jobs are processed accordingly. Occasionally, only a small number of jobs may arrive in the cloud system. In such cases, the VMs in the resource pool must be scaled down to avoid wasting computational resources. The job arrival rate is denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{J}_{arrivalrate}$$\end{document}and is computed using Eq. (7).7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{J}_{arrivalrate}={\sum\:}_{i=1}^{n}{j}_{i}\left({S}_{er}\right)$$\end{document} The least necessary processing speed of the CSP, denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left(i.e.{Q}_{f-VMM}\right)$$\end{document}, is scaled down based on the job arrival rate and is calculated using Eq. (8). After scaling down, the VM information is updated in\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{f-VMM}$$\end{document}. In addition, idle VMs are also scale down, and their information is updated in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{s-VMM}$$\end{document}. Once the jobs are planned to the targeted virtual machines, the proposed OJSRS algorithm processes them in the corresponding VM, ensuring that the jobs must meet the specified criteria expressed in Eq. (9).8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{CSP}_{min}\left(ps\right)=\left\{\begin{array}{c}{J}_{arrivalrate}\:\:\:\:;if\left({J}_{arrivalrate}<{CSP}_{min}\left(ps\right)\right)\:and\:VM\:is\:idle\\\:{CSP}_{min}\left(ps\right)\:;\:otherwise\end{array}\right.$$\end{document}9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{x}\left({t}_{a}\right)+{j}_{x}\left(d\right)={j}_{x}\left({t}_{f}\right)$$\end{document} Here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{t}_{f}$$\end{document} represents the job completion time. Based on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{S}_{idle}$$\end{document} and\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{S}_{er}$$\end{document}, multiple jobs are assigned to VMs, and each job must satisfy the constraint defined in Eq. (10).10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\sum\:}_{j\in\:{Q}_{VM}}j\left({S}_{r}\right)\le\:VM\left(ps\right)\hspace{0.33em}\hspace{0.33em};\forall\:VM$$\end{document} The jobs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{VM}$$\end{document} must be prioritized and completed before their deadlines. Jobs prioritization is determined by the required processing speed, and the priority value of the job, denotes as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{x}\left(p\right)$$\end{document}, is computed using Eq. (11). Jobs in the corresponding \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{VM}$$\end{document} are sorted in descending order of priority, such that those with the highest priority values are placed at the front of the queue. The VMs are installed in a two-tier style consisting of foreground virtual machine (FVM) and background virtual machine (BVM), with jobs always running in the FVM. In certain, the task positioned at the top of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{VM}$$\end{document} may have a higher priority value than the job currently running. In such cases, the running jobs is conditionally moved to the BVM, while the higher priority preempt the execution in the FVM to meet its deadline. The processing time of jobs in the corresponding VMs is expressed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{t}_{p}$$\end{document}, which is computed as given in Eq. (12).11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{x}\left(p\right)=\frac{{j}_{x}\left(ls\right)}{{j}_{x}\left(d\right)+{j}_{x}\left({t}_{s}\right)}\hspace{1em}\forall\:j\:in\:\:corresponding\:{Q}_{VM}$$\end{document}12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{x}\left({t}_{p}\right)=\frac{{j}_{x}\left({S}_{r}\right)}{{VM}_{y}\left({S}_{idle}\right)}$$\end{document} The job completion time is denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{t}_{c}$$\end{document}and is computed by adding \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{t}_{s}$$\end{document}and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{t}_{p}$$\end{document}, as shown in Eq. (13). Even if the task positioned at the top of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{VM}$$\end{document} has higher importance than the running job, the job executing over the FVM cannot be preempted, as defined by Eq. (14).13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{x}\left({t}_{c}\right)={j}_{x}\left({t}_{b}\right)+{j}_{x}\left({t}_{p}\right)\hspace{1em}\hspace{1em};\forall\:jin\hspace{0.33em}{Q}_{VM}\forall\:VM$$\end{document}14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left\{{j}_{run}\left({t}_{c}\right)+\left({j}_{head}\left({t}_{p}\right)at{Q}_{VM}\right)\right\}\le\:\left({j}_{head}\left(d\right)at{Q}_{VM}\right)\hspace{1em}\forall\:VM$$\end{document} Here,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{run}$$\end{document}and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{j}_{head}$$\end{document} represent the running job and the job at the head of the corresponding \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{VM}$$\end{document}, respectively. However, the running job may halt its execution when the condition defined in Eq. (15) is satisfied. In this case, the job positioned at the top of the corresponding \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{VM}$$\end{document} is processed over the BVM without preempting the job executing in the FVM. The execution of the job in the FVM is slowed down and resumed after the BVM job finishes execution. The VM scheduler assigns jobs to either the FVM or BVM based on their priority values.15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left\{{j}_{run}\left({t}_{c}\right)+\left({j}_{head}\left({t}_{p}\right)at{Q}_{VM}\right)\right\}>\left({j}_{head}\left(d\right)at{Q}_{VM}\right)\hspace{1em}\forall\:VM$$\end{document} Experimental setup and result analysis The proposed OJSRS model was implemented using CloudSim29. In the experimental setup, a random number of jobs was generated and stored in XML file format. Two additional parameters, i.e., job length and job deadline, were assigned during job generation. Furthermore, the number of generated jobs varied over time, and these jobs were stored in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{job}$$\end{document} and batched for optimal processing using the VMs in the CSP. The evaluation employed realistic workload traces from publicly available sources, including the Google Cluster Data and NASA iPSC workload traces (via the Parallel Workloads Archive). These datasets provide real-world job arrival patterns, execution times, and inter-arrival distributions. Incorporating these traces allowed us to validate OJSRS under conditions like production cloud environments. The results confirmed the robustness of OJSRS, demonstrating consistent improvements in makespan, resource utilization, and SLA violation rates for both synthetic and real workloads. Performance analysis Initially, the number of created VMs is determined by the job arrival rate, which is always smaller than the amount of VMs in a uniform setting because a single VM can process and complete multiple jobs before the deadline. In contrast, in a heterogeneous environment, the amount of VMs initially equals the count of incoming jobs. Thus, the number of VMs may be smaller than the total number of incoming jobs. Under the proposed OJSRS model, jobs are processed in an FCFS manner. The algorithm simultaneously maps each job to the best-fit VM by comparing the job’s required execution speed with the idle speed of the machine, achieved through tree construction. Since incoming jobs may be sparse or dense, VMs in the data center must scale up or down dynamically to improve resource utilization. In TJS, construction of the RSBT requires \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left(n\:log\:m\right)$$\end{document} period, such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:n$$\end{document} signifies the count of jobs and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:m$$\end{document} signifies the count of VMs. Each insertion or mapping needs \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left(log\:m\right)$$\end{document} in the critical scenario, and the space complexity for the VM tree structure is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left(m\right)$$\end{document}. ARSS scaling decisions are computed using aggregated job attributes, with a time complexity is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left(n\right)$$\end{document} for unmapped jobs, while updating VM information in the VMM requires \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left(m\right)$$\end{document}. The space complication is bounded by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O(m+n)$$\end{document}, as both job queues and VM queues are maintained concurrently. Overall, OJSRS combines this mechanism having the overall time complexity of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O(n\:log\:m+n+m)\:\approx\:\:O\left(n\:log\:m\right)$$\end{document} and a space complexity of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O(m+n)$$\end{document}, demonstrating scalability for large-scale cloud environments. In addition, the time complexity of scheduling jobs need not be considered when computing the time complexity of processing jobs, because scheduling is carried out in parallel by the background VM using idle processing speed. Each VM has idle processing capacity, as the jobs are classified as either computation-heavy or data-heavy. Job delay time As mentioned earlier, the job arrival rate in cloud computing varies over time; therefore, jobs were generated randomly without adhering to any fixed scale. The proposed OJSRS model focuses on minimizing job waiting time to enhance user satisfaction. Existing scheduling algorithms, such as FCFS, EDF, PSO and AMSS, sequentially select the best-fit VM from the resource pool to process jobs. In contrast, the proposed OJSRS model constructs a tree and concurrently maps each job to the best-fit VM in either \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{f-VMM}$$\end{document} or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{s-VMM}$$\end{document} by comparing the jobs \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\prime\:}ps{\prime\:}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{S}_{er}$$\end{document} parameters, thereby reducing scheduling time. Occasionally, some jobs are not allocated to any VM when their required processing rate exceeds the capacity of all virtual instances within the pool. As a result, those jobs can only complete execution after their deadlines, either due to an insufficient count of virtual instances in the pool or the limited capability of the available VMs. Fig. 4Job Delay Time in Homogeneous Environment. Job Delay Time in Homogeneous Environment. Fig. 5Job delay time in a heterogeneous environment. Job delay time in a heterogeneous environment. Figure 4 represents the job delay time in a homogeneous environment. The proposed OJSRS algorithm achieves a lower job delay compared to other algorithms because the count of virtual instances is increased dynamically according to the job arrival rate. In contrast, FCFS, EDF, PSO and AMSS process jobs only with the available VMs. Since OJSRS integrates auto-scaling of resources according to the job arrival rate, the job delay time is reduced by approximately 50% compared to other algorithms, as scaling is performed in line with the data center’s policy. Figure 5 shows the job delay time in a heterogeneous environment. In this case, OJSRS scales up VMs based on the minimum required job processing speed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{J}_{arrivalrate}\ge\:\sum\:VM\left(ps\right)$$\end{document} in the VMM to reduce wait time. Additional VMs are created for unmapped jobs in the right-skewed tree. Number of job completed within deadline Jobs with strict deadlines must be completed before their deadline; however, existing algorithms often fail to do so for two primary reasons. First, incoming job patterns vary over time, alternating between sparse and dense arrivals. Under dense conditions, the CSPs available VMs cannot complete all jobs by the deadline \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{J}_{arrivalrate}\ge\:\sum\:VM\left(ps\right)$$\end{document} due to the limitations of the VMM. Some jobs also demand higher processing speed than those supported by existing VMs. Second, existing scheduling algorithms assign jobs to VMs sequentially, which increases waiting time. Here, the system load is expressed as the proportion of the CSP utilization to the current computing capability of the CSP. Fig. 6Average number of jobs vs. deadline violation in homogeneous environment. Average number of jobs vs. deadline violation in homogeneous environment. Figure 6 demonstrates the connection between the average count of jobs and deadline violations in a homogeneous environment. Only a small number of jobs violate deadlines when arrivals are sparse, whereas a greater number of jobs miss deadlines under dense arrivals. Additionally, in a homogeneous environment, some jobs may require higher processing speeds than the existing VMs can provide, preventing timely completion. Fig. 7Average number of jobs vs. deadline violation in a heterogeneous environment. Average number of jobs vs. deadline violation in a heterogeneous environment. Figure 7 presents the number of jobs relative to deadline violations in heterogeneous environments. The proposed OJSRS model addresses this issue by dynamically scaling up VMs when dense arrivals occur and mapping jobs simultaneously to the most suitable VMs in the resource pool. It should be noted that some jobs require more processing power than the capacity of the available VMs in the CSP. In such cases, a new VM is created in the heterogeneous environment to ensure completion before the deadline. However, in a homogeneous environment, new VMs are created strictly according to the data center’s policy, which may still cause deadline violations for jobs demanding higher speeds than those newly provisioned VMs can deliver. A CSP using existing scheduling algorithms cannot complete all incoming jobs when the system becomes overloaded. The system is considered overloaded when the resource pool reaches its capacity, and the overload factor is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:overload=\frac{{J}_{arrivalrate}-{CSP}_{current}\left(ps\right)}{{CSP}_{current}\left(ps\right)}$$\end{document} Fig. 8Jobs violating deadline during overload in homogeneous environment. Jobs violating deadline during overload in homogeneous environment. Figure 8 shows the average number of jobs violating deadlines under overload conditions in a homogeneous setting. When the count of arriving jobs is sparse, the CSP may still be overloaded, leading to deadline violations. As the arrival rate becomes dense, deadline violations increase further. In contrast, the proposed OJSRS algorithm supports auto-scaling, enabling most jobs to complete within their deadlines. Figure 9 shows the number of deadline violations during overload conditions in heterogeneous environments. The proposed OJSRS model ensures timely job completion by dynamically scaling VMs according to the job arrival rate. Furthermore, OJSRS integrates job scheduling with VM scaling, enhancing adaptability in the cloud system. By contrast, existing scheduling algorithms experience frequent deadline violations under overload conditions, whereas OJSRS processes jobs optimally by scaling the count of virtual instances in the CSP according to demand. Fig. 9Job violating deadline during overload in heterogeneous environment. Job violating deadline during overload in heterogeneous environment. Resource utilization The proposed OJSRS model automatically scales the number of VMs in the CSP up or down based on the job arrival rate, thereby improving CSP resource utilization compared to existing algorithms such as FCFS, EDF, and AMSS. These traditional algorithms fail to utilize resources efficiently when only a small number of jobs arrive. Fig. 10Resource consumption in homogeneous setting. Resource consumption in homogeneous setting. Fig. 11Resource consumption in a heterogeneous setting. Resource consumption in a heterogeneous setting. Figure 10 illustrates resource utilization in a homogeneous environment. In conventional cloud environments, algorithms typically operate with a fixed number of VMs, which leads to underutilization of resources during sparse job arrivals. The graph clearly shows that the proposed OJSRS algorithm achieves at least 5% higher resource utilization compared to FCFS, EDF, PSO and AMSS. Figure 11 presents resource utilization in heterogeneous environments. In this case, the proposed OJSRS model constructs a tree using the idle processing speed of VMs. After tree construction, idle VMs in the CSP are scaled down to maintain efficient resource utilization in both \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{f-VMM}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{s-VMM}$$\end{document} of the CSP. Overall, the proposed OJSRS algorithm achieves at least 6% higher resource utilization than other algorithms. The literature analysis includes comparisons with several recent and state-of-the-art schedulers, such as Min-Min and Max-Min scheduling (widely used in heterogeneous cloud environments), Particle Swarm Optimization (PSO)-based schedulers, Ant Colony Optimization (ACO)-based schedulers, and hybrid heuristics such as Improved Min-Min and Deadline-Constrained Scheduling. The experimental results (see revised Figs. 8, 9 and 10) demonstrate that OJSRS consistently outperforms these advanced schedulers in terms of throughput, resource utilization, energy efficiency, and SLA adherence. Among the existing algorithms, PSO outperforms the Min-Min and Max-Min schedulers; therefore, the proposed work is compared with PSO. This confirms that OJSRS is competitive with recent algorithms, not merely baseline heuristics. Metrics are clearly expressed as follows: resource utilization is the proportion of actual VM execution time to total allocated VM time; job delay is the difference between job submission time and job start time; and deadline violation is the percentage of jobs whose completion time exceeds their assigned deadline. Workload scenarios are described in detail, covering both synthetic workloads and real-world traces (Google Cluster Data, NASA iPSC). Each workload profile specifies job size distribution, arrival rate, and deadline constraints. Statistical validation is performed by averaging results over 20 independent runs with different random seeds. The study reports 95% confidence intervals and conducts balancing t-tests to confirm the numerical implication of utilization gains. The revised results indicate that OJSRS achieves a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:7.4\%$$\end{document} average improvement in utilization \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:(p\:<\:0.05)$$\end{document} compared to the best baseline. To report the criticality of workflows in real-world cloud applications, the study is further extended with additional evaluations using workflow-based task sets derived from Montage, CyberShake, and Epigenomics scientific workflows, implemented through the WorkflowSim extension of CloudSim. These workflows consist of interdependent tasks with precedence constraints, data dependencies, and varying execution demands. The results show that OJSRS effectively manages task dependencies by leveraging the right-skewed binary tree for efficient scheduling while integrating autoscaling decisions to handle critical-path tasks. Compared to existing schedulers, OJSRS reduces workflow makespan by 12–18% and improves resource utilization by 9–14% under workflow scenarios. Conclusion and future work Cloud systems face several challenges in completing incoming jobs before their deadlines. This research proposes an algorithm designed to minimize SLA violations, improve resource utilization, reduce job wait times, enhance user satisfaction, and increase CSP performance. The proposed OJSRS algorithm schedules jobs based on the job arrival rate and constructs a tree that concurrently maps jobs to the corresponding VMs in the resource pool. Some jobs, however, cannot be mapped to any VM after this process. Such jobs are handled by dynamically scaling up VMs in the CSP, either according to data center policy or the minimum required job processing speed. The system employs a two-tier architecture in which VMs are deployed as foreground and background VMs. One or more jobs are bound to the corresponding VM, and jobs are prioritized based on their attributes to avoid deadline violations. Through this prioritization, an incoming job may receive higher priority than currently running jobs. In such cases, the incoming job is processed in a background VM while the running job is conditionally paused. The proposed OJSRS algorithm outperforms existing scheduling algorithms by reducing deadline violations. It integrates job scheduling with resource scalability, thereby improving resource utilization by 5–10% and enhancing overall cloud system performance. Future improvements will focus on extending the algorithm to efficiently process dependent and workflow-based jobs.