Reliability-Aware Cloud Task Scheduling using Large Neighborhood Search Optimization

Task scheduling is a fundamental challenge in cloud computing, directly affecting system balance, reliability, and service quality under large-scale and dynamic workloads. The primary objective of this study is to address load imbalance and execution failures in cloud environments by proposing an efficient task scheduling framework based on the Large Neighborhood Search (LNS) algorithm. The proposed approach is implemented using the CloudSim simulator and evaluated on realistic Alibaba workload traces, which capture heterogeneous and time-varying cloud task characteristics. LNS systematically improves initial scheduling solutions by selectively destroying and repairing large portions of the task-VM mapping, enabling effective exploration of the search space while avoiding premature convergence. The performance of the proposed scheduler is assessed using key metrics, including degree of imbalance, VM success rate, and task failure rate, and is compared against well-established metaheuristic algorithms such as Genetic Algorithm (GA), Ant Colony Optimization (ACO), and Particle Swarm Optimization (PSO). Experimental results demonstrate that the LNS-based scheduler reduces the degree of imbalance by up to 36.8%, improves VM success rate by approximately 13.1%, and decreases task failure rate by nearly 42.2% compared to the benchmark algorithms. These improvements highlight the robustness and adaptability of LNS in handling large-scale cloud workloads. The findings indicate that the proposed LNS-based scheduling framework can significantly enhance load balancing and reliability in cloud data centers, making it a promising solution for efficient and fault-aware cloud resource management.