Dynamic Server Provisioning for Carbon-Neutral Data Centers

In light of the growing trend of data center carbon emission that has raised serious sustainability concerns, data center operators are aggressively seeking ways to minimize the overall energy consumption by using more on-site renewable energy and ultimately achieving carbon neutrality (or "net"). In this paper, we propose a dynamic server provisioning algorithm, called SPAN (Server Provisioning for carbon Neutrality), to control the number of active servers for minimizing the data center operational cost (defined as a weighted sum of electricity cost and delay cost) while achieving carbon neutrality without requiring long-term future information. Leveraging the recently-developed Lyapunov optimization technique, it is rigorously proved that SPAN achieves a close-to-minimum operational cost compared to the optimal offline algorithm with future information, while bounding the potential violation of carbon neutrality. The simulation result is also consistent with our analysis, showing that SPAN can reduce the average operational cost by 20% while maintaining a carbon neutral data center.