A resource and deadline-aware dynamic load balancing scheduler for cloud jobs

Cloud Service Providers (CSPs) deliver computing resources to users under a pay-as-you-go model, where timely task execution and efficient resource utilization are essential for maintaining Service Level Agreements (SLAs) and ensuring customer satisfaction. However, many existing cloud job scheduling algorithms fail to simultaneously satisfy job deadline constraints and achieve high resource utilization, primarily due to insufficient deadline awareness and ineffective workload distribution in heterogeneous cloud environments. To address these challenges, this paper proposes a resource- and deadline-aware dynamic load balancing scheduling algorithm (DBS) for cloud computing systems, supported by a formal problem formulation and a polynomial-time complexity analysis. The proposed approach allocates incoming jobs to heterogeneous virtual machines (VMs) by jointly considering job computational requirements, VM processing capabilities, and deadline feasibility. Furthermore, the scheduler dynamically prioritizes urgent jobs through deadline-based queue sorting and intelligently repositions newly arrived tasks using a Suitable Position Finder (SPF) mechanism to minimize task rejection and deadline violations. Extensive simulations were conducted using the realistic GoCJ benchmark dataset across four workload categories: small and medium, large, extra-large and huge, and combined heterogeneous workloads. All results are reported over 30 independent simulation runs and validated using the Wilcoxon signed-rank test ( \(p < 0.05\) $p<0.05$ ). The experimental results demonstrate that DBS consistently outperforms state-of-the-art algorithms in terms of multiple performance metrics. Specifically, DBS achieves up to a 3.22 \(\times \) $\times$ reduction in task rejection ratio, a 93.6% improvement in average resource utilization compared to classical heuristic-based approaches, a 1.92 \(\times \) $\times$ increase in throughput, a 40.54% improvement in overall gain, and a 1.35 \(\times \) $\times$ reduction in makespan. These results confirm that the proposed scheduler effectively enhances deadline compliance, workload balance, scalability, and overall cloud system performance under heterogeneous and high-load conditions.