Active Elastic Scaling Strategy Based on Spatio-Temporal Graph Neural Network

The microservice architecture has accelerated the development of containerized deployment in cloud computing, with elastic scaling emerging as a key technology to address dynamic workloads. Traditional reactive strategies struggle to adapt to complex load fluctuations due to cold start latency and rigid resource allocation. Although proactive strategies have gained attention, their prediction accuracy remains challenging, especially as they often overlook the invocation relationships among microservices. To address this, this paper proposes a load prediction method based on a spatio-temporal graph neural network, which integrates Graph Attention Networks (GAT) to model spatial dependencies between microservices and Gated Recurrent Units (GRU) to capture temporal dynamics. Experimental results demonstrate that the proposed method outperforms traditional models in prediction accuracy.