Students from the EDISS programme have had their research paper accepted at the International Conference on Software Architecture (ICSA) 2026, the premier international venue for practitioners and researchers working on software architecture, component-based software engineering, and software quality attributes and their architectural implications.
The paper, titled “Green Autoscaler for Performance Aware Microservices: a Machine Learning Approach,” is co-authored by Thanh-Phuc Tran, Abhinandan Roul, Ishara Galbokka Hewage, and Mahira Joytu. The work was jointly co-supervised by researchers from Åbo Akademi University and the University of L’Aquila: Roberta Capuano, Eoan O’Dea, Rafiullah Omar, Sébastien Lafond, Hergys Rexha, and Henry Muccini, reflecting a strong international and interdisciplinary collaboration.
Carbon-Aware Software Architecture for Cloud-Native Systems
As cloud-native systems increasingly adopt microservice architectures, autoscaling mechanisms play a crucial role in maintaining performance under dynamic workloads. However, widely used solutions—such as the Kubernetes Horizontal Pod Autoscaler (HPA)—primarily optimize performance metrics and overlook the environmental impact of scaling decisions, despite rising concerns over energy consumption and carbon emissions.
To address this gap, the authors propose a carbon-aware autoscaling system that integrates sustainability concerns directly into architectural decision-making. The approach leverages Spatio-Temporal Graph Convolutional Networks (ST-GCNs) to jointly model workload dynamics and inter-service dependencies, while incorporating real-time regional carbon intensity data. By introducing carbon-aware thresholds, the autoscaler dynamically balances performance requirements with environmental impact.
Promising Results for Sustainable Cloud Architectures
The proposed approach was evaluated on three benchmark microservice applications, using a synchronized monitoring infrastructure capturing performance, energy, and carbon metrics. Experimental results demonstrate that the system can significantly reduce emissions while preserving application performance:
- ~24% average carbon emission reduction in high-carbon-intensity regions (300 gCO₂/kWh)
- ~17% reduction in low-carbon-intensity regions (100 gCO₂/kWh)
- Comparable performance to traditional autoscaling in low-carbon conditions
These results highlight the potential of machine learning–driven, carbon-aware architectural solutions to support greener cloud infrastructures without sacrificing performance guarantees.
The acceptance of this paper at ICSA 2026 underlines the strong contribution of EDISS students to state-of-the-art research in software architecture, sustainability, and cloud systems, and reinforces the programme’s emphasis on addressing global challenges through advanced software engineering research.
Last updated on 8 February 2026
