- With the rapid growth of microservices-based applications in the cloud, the demand for scalable computing resources has led to increased energy consumption and carbon emissions.
- Traditional approaches that limit computing resources during high carbon intensity periods often compromise application performance, particularly response times.
- This project aims to build and evaluate a machine learning-based carbon-aware autoscaler that addresses the tradeoff between sustainability (energy consumption) and performance ().
- Incorporate Carbon Awareness: Use carbon intensity data to guide scaling decisions, aligning cloud operations with sustainability goals.
- Balance Performance & Energy Trade-off: Maintain critical performance metrics (e.g., response time) while reducing carbon emissions.
- Integrate Machine Learning: Explore ML-based autoscaling techniques to enhance energy efficiency without significant performance degradation. This solution aims to improve the sustainability of cloud-native applications while ensuring optimal performance. 🚀
-
Inspired by the DeepScaler [https://github.com/SYSU-Workflow-Lab/DeepScaler] project, we have developed a novel mechanism to take into account the carbon emission in scaling decisions, and make better scaling decisions which are carbon-aware and have least impact to the application response time.
-
Proposed Pipeline
- Model Architecture
- Response Time comparison
- Carbon footprint comparison
