RARE: Renewable Energy Aware Resource Management in Datacenters

AbstractThe exponential growth in demand for digital services drives massive datacenter energy consumption and negative environmental impacts. Promoting sustainable solutions to pressing infrastructure challenges is crucial. Several hyperscale cloud providers have announced plans power their datacenters using renewable energy. However, integrating renewables the challenging because generation intermittent, necessitating approaches tackle supply variability. Hand engineering domain-specific heuristics-based schedulers meet specific objective functions such complex dynamic green environments time-consuming, expensive, requires extensive tuning by domain experts. The need smart systems system software employ multiple sources (wind solar) intelligently adapting computing generation. We present RARE (Renewable Aware Resource management), a Deep Reinforcement Learning (DRL) job scheduler that automatically learns effective scheduling policies while continually datacenters’ environment. resulting DRL performs better than heuristic with different workloads adapts intermittent from renewables. demonstrate design parameters that, when tuned correctly, produce performance. Finally, we can learn improve upon existing Offline Learning.KeywordsRenewable energyDatacentersJob schedulingDeep reinforcement learning