As cloud computing continues to scale, optimizing resource usage and efficiency is essential for reducing both costs and environmental impact.
While Machine Learning (ML) is often considered the go-to solution for managing complex systems and predicting future resource demand, its use isn’t always necessary.
The deployment of ML models introduces challenges, including significant overheads, interpretability issues, and sustainability concerns.
Our research shows that cloud resource usage data exhibits strong temporal correlations, enabling lightweight heuristics to accurately forecast future demand.
This insight suggests that simpler, data-driven approaches can be just as effective as ML, underlining the importance of using ML judiciously.
In conclusion, this talk emphasizes the need for a smart, low-cost, and low-carbon approach to cloud resource management, where ML is used only when truly beneficial.
