Carbon-Aware Load Balance Control of Data Centers With Renewable Generations

With the increasing environmental issues, the energy consumption and carbon emissions of data centers have become a major concern. However, in previous works, the cost and carbon reduction potential of geographically dispersed data centers with renewable generations is not fully explored due to the difficulty in matching renewable generations with stochastic incoming jobs. In this paper, the temporal and spatial variability of the carbon footprint and electricity price is revealed. Integrated with the distributed characteristics of renewable generations and geographically dispersed feature of the data centers, the carbon emission reduction potential of data centers with renewable generations is explored, leading to triple uncertainties in electricity price, fuel mix and renewable generation. To navigate such a potential, a virtual queue algorithm is designed, which makes online strategies for job scheduling of the data centers. By introducing historical correction terms, high-precision matching of the workload and renewable generation can be achieved with triple uncertainties. This leads to the economic and environmental friendliness of the proposed mechanism, which can achieve $O(\sqrt{T})$ expected regret and constraint violation. Simulations based on real-world data from several states of Australia demonstrate the effectiveness of the proposed framework in cost and carbon emission reduction with triple uncertainties.