Cloud-native microservice systems increasingly rely on autoscaling to maintain performance under fluctuating workloads, yet scaling decisions strongly influence energy consumption and carbon emissions, making sustainability a growing concern for modern cloud infrastructures. Traditional mechanisms such as the Kubernetes Horizontal Pod Autoscaler optimize only performance metrics and ignore the carbon intensity of electricity sources, leading to excessive provisioning and higher emissions. To address this limitation, this paper proposes a Carbon-Aware Autoscaling System based on Spatio-Temporal Graph Convolutional Networks that jointly model workload dynamics and inter-service dependencies while integrating real-time regional carbon intensity. The autoscaler dynamically moderates scaling levels using carbon-aware thresholds, enabling adaptive tradeoffs between performance and sustainability. We evaluated our approach on three benchmark microservice applications using a synchronized monitoring stack for performance, energy, and carbon metrics. Experimental results show that the proposed approach achieves an average carbon emission reduction of approximately 24% in high-intensity regions (300 gCO2/kWh) and 17% in low-intensity regions (100 gCO2/kWh) as compared to HPA while maintaining comparable performance in low-carbon conditions.

As the integration of Machine Learning (ML) models becomes pervasive in software systems, the associated energy costs have emerged as a critical concern. This research delves into the energy efficiency of ML-enabled systems, focusing on the ML component itself and its impact on the overall system. Our primary emphasis is on the data-centric approach, particularly in the context of feature selection and handling concept drift, and how these energy-efficient components affect the overall energy consumption of ML-enabled systems. In our initial investigation, we explored feature selection methods and identified significant variations in their energy consumption. This led us to delve deeper into understanding how different techniques for scoring features contribute to the overall energy footprint of ML models. Subsequently, we are examining the impact of changes in data distribution, often referred to as concept drift, on model accuracy and the associated energy costs. Our empirical experiments will reveal insights into energy-efficient strategies for handling concept drift, a crucial aspect of maintaining ML-enabled systems. We will compare various methods and their effectiveness in mitigating the adverse effects of concept drift while keeping energy consumption in check. The findings from our research contribute to the development of sustainable and energy-efficient ML models within the broader context of software engineering. Lastly, we will compare how different alternatives of ML components in ML-enabled systems affect the overall energy consumption of ML-enabled systems.