Adaptive green cloud applications: Balancing emissions, revenue, and user experience through approximate computing

Abstract

Organisations will soon be required to take an active role in the green transition by minimising the environmental impact of their operations, including emissions from their information systems. National and international regulations are expected to drive this shift by enforcing carbon budgets that organisations must comply with. As a result, applications must not only be aware of their carbon footprint but also operate within these budgetary constraints.

Traditional methods, such as time and location shifting, have been used to mitigate emissions, but their impact is limited and not applicable to all types of applications. Recent research suggests that reducing an application’s environmental footprint can be achieved through approximation techniques, where workflows dynamically adjust at runtime by scaling back certain functionalities or features. However, this approach introduces trade-offs: limiting functionalities can reduce revenue, especially when tied to third-party agreements, and may also degrade the user experience. Thus, striking a balance between carbon reduction, business objectives, and user satisfaction is crucial.

We present a carbon-aware application management approach that leverages approximate computing techniques to balance sustainability, user experience, and revenue. Our method dynamically optimises the configuration and scaling of individual software components within a predefined carbon budget. Through simulation-based evaluation across diverse regions, carbon budgets, and application setups, we demonstrate that the approach effectively adapts to fluctuating workloads and regional variations in carbon intensity.