Cloud platform resource capacity management and scheduling optimization based on data graph
Ziliang Qiu, Wei Deng, Juzheng Huang, X F Liu, Binhua Ren, Siyuan Qin · 2025
This paper proposes a data graph based cloud platform resource capacity management and scheduling optimization method to address the difficulty of handling large-scale multi-dimensional constraints in cloud platform resource scheduling. We constructs a multi-level resource graph model for cloud platforms, which realizes the formal description of resource entities, attributes and relationships. Based on this, we designe a resource capacity prediction method that integrates graph neural networks, which improves prediction accuracy by modeling node feature sequences. Meanwhile, we propose an improved multi-objective optimization algorithm that maximizes resource utilization, minimizes system energy consumption, and balances load as optimization objectives, ultimately achieving adaptive optimization of scheduling schemes. The experimental results show that compared with traditional methods, the proposed method improves resource utilization by 24%, reduces average response time by 52.8%, and achieves an SLA satisfaction rate of 98%. Further, in large-scale scenarios (1000+containers), the algorithm can converge to a stable solution within 9 iterations. When the cluster size expanded from 10 nodes to 50 nodes, the scheduling decision time only increased by 42%, verifying the good scalability of this method. The research results have important theoretical and practical value for improving the efficiency of cloud platform resource management.