Servers and data centers energy performance metrics

Abstract

Abstract A thermodynamic approach for evaluating energy performance (productivity) of information technology (IT) servers and data centers is presented. This approach is based on the first law efficiency to deliver energy performance metrics defined as the ratio of the useful work output (server utilization) to the total energy expanded to support the corresponding computational work. These energy performance metrics will facilitate proper energy evaluation and can be used as indicators to rank and classify IT systems and data centers regardless of their size, capacity or physical location. The current approach utilizes relevant and readily available information such as the total facility power, the servers’ idle power, the average servers’ utilization, the cooling power and the total IT equipment power. Experimental simulations and analysis are presented for a single and a dual-core IT server, and similar analysis is extended to a hypothetical data center. The current results show that the server energy efficiency increases with increasing CPU utilization and is higher for a multi-processor server than for a single-processor server. This is also true at the data center level however with a lower relative performance indicator value than for the server level.