Analyzing Power Usage in Datacenter Workloads Using Random Forest and LSTM Models

Datacenters, crucial in modern computing, often face power management inefficiencies. This study addresses this by analyzing workload power consumption and using a Random Forest model. It comprises two phases: data collection and predictive modeling. In data collection, we gathered a comprehensive dataset covering CPU, GPU, memory usage, total power consumption, and user request types. This data trained a Random Forest model, achieving high accuracy with an RMSE of 0.016 in predicting power consumption patterns. The predictive modeling phase focused on specific request types: computing, collaborating, streaming, and Others. Leveraging Time Series Models, we made precise power consumption predictions for each type, revealing critical high usage periods. This empowers operators to make informed decisions during peak demand. Our findings highlight the potential for optimizing energy usage in datacenters. Accurate prediction and identification of critical moments enable real-time energy minimization, enhancing efficiency and sustainability. This research fills a gap in datacenter energy optimization, promising practical benefits for operators and the environment, and paving the way for future research in datacenter management.