Power consumption analysis for data centers with independent setup times and threshold controls

Abstract

Data center consumes a large amount of energy since it has a huge number of servers. Therefore, a reduction of a few percent of the power consumption means saving a large amount of money and the environment. In the current technology, an idle server still consumes about 65% of its peak. Thus, an efficient way to save energy is the shutdown of server when the workload is low. Mitrani [3] proposes a threshold control policy where a block of servers is turned on and off according to two thresholds for the number of jobs present in the system. Relaxing the assumption in Mitrani [1] that after a setup time, all the servers in the block are active concurrently, we consider a more realistic model where each server has an independent setup time. We formulate the system by a two dimensional Markov chain and derive performance measures using a generating function approach. The results allow us to investigate the performance and energy consumption trade-off.