XDL-IRTS: an Intelligent and Resilient Task Scheduling in Large-Scale Cloud Environments

Large-scale cloud computing environments require task scheduling mechanisms that simultaneously ensure high performance, strong reliability, and transparent decisionmaking under dynamic and failure-prone workloads. The primary objective of this work is to design an intelligent, failureaware, and explainable scheduling framework that minimizes virtual machine (VM) failures while maximizing task success rate and system throughput. To achieve this objective, this paper proposes an Explainable Deep Learning-based Intelligent and Resilient Task Scheduling (XDL-IRTS) framework that integrates workload and reliability feature learning with an explainable deep decision model and a resilience-driven scheduling strategy. The proposed approach proactively avoids failure-prone resources while adapting scheduling decisions to evolving workload conditions. XDL-IRTS is implemented using the CloudSim simulator and evaluated under realistic Google Cloud workload traces, and its performance is systematically compared with Genetic Algorithm (GA), Ant Colony Optimization (ACO), and Particle Swarm Optimization (PSO). Experimental results demonstrate that the proposed framework achieves a$\mathbf{3 0 - 3 5 \%}$reduction in VM failure rate, an$\mathbf{8 - 1 5 \%}$improvement in task success rate, and a$\mathbf{1 0}-\mathbf{2 5 \%}$increase in throughput across different workload scales. Furthermore, the integrated explainability mechanism provides transparent insights into scheduling decisions, enhancing operational trust and practical applicability in large-scale cloud environments.