Design and Analysis of Job Assignment in Server Farms of Data Centers

We study the job assignment policy in a data center of diverse processor-sharing servers, and each server has different service rates, energy consumption rates, and buffer sizes. We came up with a job assignment algorithm, called Energy -efficiency available server plus Regularization term takes up Idle energy (ERAIP), that reduces the mean service time while optimizing the energy efficiency. According to Kolmogorov equations which describe the trend of transitions between system states, we prove the optimized performance of a system with two servers. Experiments show that ERAIP can improve energy efficiency while reducing the mean service time.