resources

This paper illustrates how today’s data center workload is managed and optimized through the presentation of several real-use scenarios. Recognizing that servers deployed into any environment—office, enterprise data center, or cloud data center—are sized to handle a specified workload or set of workloads, we can assess the efficacy of idle power and active efficiency (defined as a weighted geomean[1] active efficiency or active efficiency metric) metrics as a measure of server energy efficiency using the deployed power methodology. The deployed power methodology uses the measured, aggregated SERT performance and power data to determine the number of servers required to meet a large target workload and the data center power use of that group of servers. Comparing the deployed power and active efficiency of servers that pass the idle test with the servers that pass the active efficiency test demonstrates that the use of idle power as a policy tool for regulating server energy efficiency will unfairly exclude higher performance servers from the market. Removing these higher power, higher performing servers from the market could result in over 30 percent higher data center energy consumption (Figure 12) than would occur if an active efficiency metric were used. Furthermore, since the server idle power trend is expected to remain flat or slightly worsen, subsequent tightening of any idle limit will further deplete the number of higher performing servers, resulting in a lost opportunity to minimize data center energy consumption. Based on the robust data presented, including future data center market trends and year-over-year server performance improvement trends, The Green Grid strongly advocates moving away from the server idle power approach and toward the weighted geomean active efficiency metric based on SERT. The Green Grid looks forward to constructive engagement with the stakeholders and the advocacy community.

[1] ‘geomean’ is an abbreviated term that means the same as ‘geometric mean’ used in statistics to describe one of the averaging methods for a dataset.