Small Embedded Data Centers: High energy use suggests high potential for savings
CEE’s research team explores and identifies cost-effective energy efficiency measures to improve utility programs and related policy efforts. We conduct field research in buildings ranging from homes to high-rise office towers, analyzing valuable data on the energy savings and performance characteristics of energy upgrades. Among our recent research projects, CEE’s team has been looking at the operational efficiency of small data centers in commercial and industrial buildings.
Small embedded data centers (SEDCs) are among the fastest growing users of electrical energy in commercial buildings. Through field studies and stakeholder engagement, this project gathers information to improve the operational efficiency of SEDCs in commercial and industrial settings.
A server is any computer whose core purpose is to provide functionality for other computers or devices, often acting as a home base for data storage or providing centralized capacity for processing, typically drawing lots of energy to run as smoothly as possible. “Server virtualization” uses software to consolidate multiple virtual servers on a single physical server host, reducing the number of actual physical servers needed while cutting overall energy use.
When CEE’s project was initially proposed, the project team expected to identify server virtualization as an important energy efficiency measure for SEDCs. But in the time between the project proposal and the start of the study, server virtualization has become much more common, and all but one of the data centers in this study already employ some level of server virtualization.
Consequently, the team has focused more on identifying operation measures, and monitoring data center energy use has revealed some unique opportunities.
The ins and outs of data collection
The project team is monitoring a total of 23 SEDCs at 11 participant sites, including two architectural design firms, a healthcare clinic, three small industrial companies, two public school districts, a law firm, a municipal city hall, and a nonprofit. There are two non-obtrusive approaches to monitoring and data collection, depending on how the uninterruptible power supply (UPS) is connected. For each approach, data is collected at one-minute intervals.
The first approach is from Packet Power, a Twin Cities company that manufactures affordable, commercially available equipment to monitor power usage in data centers. Packet Power cables are used when the data center’s UPS is plugged into the wall, attached between the wall source and the UPS. Each UPS has a battery that keeps the power on during the short time that the UPS is disconnected from the wall, so this method of monitoring prevents server downtime during installation and removal.
The second approach uses current transformers (CTs), a monitoring system that was developed specifically to take measurements at the electrical panel. This approach is used when the UPS is connected directly to the electrical panel, and it can also be used to measure the power to the computer room air conditioners (CRACs). The CTs clip over the cables in the panel and do not require any interruption of service. However, a licensed electrician must perform the electrical panel work.
Preliminary results — how one SEDC is doing
The project team has begun to examine the energy use of some participant sites. The power use of one UPS in the nonprofit data center for the month of May 2016 is shown below:
The power load for this site is about 1.7kW and, while there is some fluctuation, it stays around this level 24 hours a day, seven days a week. Normal business hours for this office are 8 am to 5 pm, Monday through Friday. However, during nighttime hours and weekends the data center is still consuming a small amount of energy from activities such as scheduled backups and updates and a few employees who access the system remotely. Looking at the data, there is an opportunity to reduce the power demand for over 70% of the work week through options that are similar to standby or sleep modes. These options include:
Virtual machine migration to further reduce the number of physical servers that are in use during off peak hours;
Energy management scheduling of the network switches into standby mode; and
Possible sleep modes for storage arrays.
There are also opportunities in this data center for better airflow management to optimize the delivery of cooled air and the removal of hot air from the room. More efficient cooling would allow CRACs to be resized and setpoint temperatures to be raised. As with others in the study, the room where this data center resides was not originally designed for its current purpose. During the winter the building’s HVAC ductwork often continues to heat the space, working at odds with high-energy efforts to cool it.
Most SEDCs have equipment refresh rates of about five years, which limits the opportunities to improve energy efficiency through equipment replacement. In addition, replacing equipment can be expensive and is often not cost effective. Fortunately, preliminary results from this study indicate that solid opportunities to improve efficiency also lie in scheduling and operations adjustments.
As the research continues, CEE will continue to identify operational measures to reduce energy use and assess the energy saving potential. Stay tuned for additional early findings in the coming months.
This project was supported in part by a grant from the Minnesota Department of Commerce, Division of Energy Resources, through the Conservation Applied Research and Development (CARD) program. And with co-funding by CEE in support of its nonprofit mission to advance research, knowledge dissemination, and program design in the field of energy efficiency.