Issue link: https://cp.revolio.com/i/809568
THE DATA CENTER JOURNAL | 15 www.datacenterjournal.com performing the conversion consumes energy and dissipates it in the form of heat. In considering how much energy is lost, let's look at the last conversion step. An IT server typically requires about 735 W of power. Ignoring losses in the power supply, the actual required power for the processing tasks is 575 W. at's a loss of 160 W, as Figure 2 shows. Of course, the server has additional internal losses—about 15 W—but we will ignore them. At $0.20 per kilowatt (kW) per hour (h), the year's loss translates to 0.160 kW x $.20/kW/h x 8,760 h = $280.32 per server. Assuming five servers per rack, the year cost is $1,401.60 per rack. IT managers ordering servers and other equipment have the option of ordering without the power supply; the equipment then requires a DC input, typically at 12 V. Since the IT loads operate at DC voltage, wouldn't it be ideal if no conversions were necessary? Would it translate to no losses? Let's examine this possibility in more detail. Telecom operators are required to maintain telephone lines using DC power and use DC power plants in that can provide up to 24 hours of battery backup. In this environ- ment, only the initial conversion shown in Figure 1 is occurring. So why not use the same configuration in data centers? Because by standard established in the early de- velopment of landline phones, telecom infrastructures use a low voltage of –48 volts. For given power requirements, low voltage means high currents. High-current distribu- tion requires thick and costly copper cabling to distribute low DC voltage directly to the loads. Furthermore, large copper cables introduce resistance, which would decrease the voltage arriving at the load. Resolving these obstacles have been proven to be too costly. Some telecom provid- ers, however, have opted to use their existing DC-plant power to support their switch rooms and to apply invert- ers to support their data centers. In this scenario, the third conversion shown in Figure 1 is unnecessary. ere have been recent discussions regarding the use of 380VDC for distributing power in a data center. is approach would resolve the low-voltage obstacle, enabling the installation of smaller cabling. But several new techni- cal obstacles arise. A double-conversion UPS (conversions 1 and 2 in the first figure) implements a bypass such that a failure sends the input directly to the output, avoiding the interrup- tion of power to the load. In the 380 VDC scenario, only conversion 1 is required. In the event of a rectifier failure, the load is solely dependent on the battery, which is fine if the battery has extended autonomy—say, 24 hours. is capability would give a technician on call only that amount of time to repair the rectifier. Otherwise, the load must shut down to allow repair. e other obstacle is recharge time for batteries. e higher the battery voltage, the higher the power require- ments for recharging. e rule of thumb for recharge time is 10 times the discharge time. A 12-minute battery system that has completely discharged would thus take two hours to recharge. To recharge a 24-hour battery system, installation of additional rectifiers exclusively dedicated to recharging batteries would be necessary. concluSIon e advancement of battery technologies and in- creased use of alternative energies will aid in minimizing AC power loss in a data center. Demand for lower energy consumption is further met by the construction of hyper- scale data centers. Agencies are establishing standards, manufacturers are offering servers with the appropriate voltage inputs and UPS suppliers are developing DC-based products. I therefore believe we are three years from global implementation. n about the author: Yves Bouhadana is the Sales manager, north America. riello uPS y.bouhadana@riello-ups.com Figure 2. Power loss in a typical server.