Counterfactual Autoscaling for Resource-Efficient Service Orchestration in the Cloud–Edge Continuum

Abstract

Cloud–edge computing enables scalable and resilient deployment of microservice-based applications, however achieving resource efficiency while ensuring stringent Quality of Service (QoS) remains challenging. The strong interdependencies among microservices and non-linear latency effects near resource saturation render conventional workload-driven autoscaling ineffective in complex distributed environments. This paper introduces CARSO (Counterfactual Autoscaling and Resource-efficient Service Orchestration), a proactive and interpretable framework that integrates eXplainable Artificial Intelligence (XAI) into the autoscaling process. CARSO employs counterfactual reasoning to derive minimal resource adjustments that proactively prevent QoS violations. The framework includes two core components: i) a Counterfactual Vertical Autoscaling (CVA) scheme that anticipates and mitigates performance degradation and ii) a Latency-Aware Resource Orchestration (LARO) policy that coordinates scaling and placement actions to balance resource efficiency and end-to-end latency across the cloud–edge continuum. Extensive experiments demonstrate that CARSO outperforms state-of-the-art proactive autoscaling frameworks in both QoS compliance and overall resource utilization.