This paper evaluates the impact of workload distribution strategies on energy consumption and system stability in data centers using real-world workload traces derived from Google cluster data. A nonlinear power model is employed to capture realistic system behavior under varying CPU utilization levels. Multiple workload strategies, including smoothing, consolidation, and threshold limiting, are analyzed to assess their effectiveness in reducing energy consumption and improving system stability. The dataset consists of over 400,000 processed samples, enabling a comprehensive evaluation of workload behavior over time. Experimental results show that while total energy reduction across strategies is relatively small, significant improvements can be achieved in reducing peak power and system variability. In particular, threshold limiting demonstrates the most effective performance in controlling peak loads and enhancing system stability. These findings highlight the importance of considering both energy efficiency and stability metrics when designing data center optimization strategies, rather than focusing solely on total energy consumption.
Singh et al. (Thu,) studied this question.