QMScaler: A QoS-Constrained Resource-Efficient Microservice Autoscaling Framework for Edge Environments

Abstract

Edge computing has become a key paradigm for supporting low-latency and high-concurrency services deployed with microservice architectures. However, limited device resources, highly dynamic workloads, and complex service dependencies make it difficult for existing autoscaling approaches—often relying on static thresholds or simple workload prediction—to ensure both QoS guarantees and efficient resource utilization. In particular, most prior studies overlook the QoS impact of instance migration paths during scaling and fail to account for the heterogeneous contributions of microservices to end-to-end latency, resulting in unbalanced resource allocation and degraded overall performance. To address these challenges, we propose QMScaler, a QoS-driven microservice horizontal scaling framework based on Monte Carlo Tree Search (MCTS). QMScaler aims to minimize the number of container instances while meeting the QoS requirements of multiple application functions. It incorporates a fine-grained performance model that captures the effects of scaling actions and migration paths, a microservice importance metric that prioritizes resources for critical services, and a domain-knowledge-guided heuristic MCTS algorithm that improves search efficiency and decision stability. Experiments on real edge clusters demonstrate that QMScaler achieves required QoS with fewer instances and exhibits superior adaptability and robustness under dynamic and bursty workloads compared to existing methods.