We are more familiar with heat generation in electronic gadgets and methods followed for its removal. Heat sinks are commonly recognized for their function – removing unwanted heat by conduction and convection. Most engineers know how to keep the temperature of the components on their printed circuit boards such as ICs below some maximum allowable value. They are also aware of the heat within the overall enclosure, which may be for a standard rack of boards, a power supply, or a DVR.
Engineers follow several techniques to remove heat from ICs, boards, or enclosures. These often involve the use of one or more heat sinks, heat pipes, heat spreaders, cold plates, and fans. Methods that are more sophisticated use active cooling approaches that include air conditioning or cooling with liquid flowing through embedded pipes. All these are good techniques for handling heat generated with a few kilowatts of power.
However, things change when megawatts of power is involved. Consider for instance, a hyperscale data center that offers a massively scalable compute architecture. Usually made up of small, individual servers called nodes, the hyperscale data center provides computing power, storage, and networking, with the nodes clustered together and managed to form a single entity. Inexpensive, off-the-shelf servers form the nodes. As demand increases, more nodes are attached. Although no formal standard is available as to the minimum power dissipation that can be considered hyperscale, it is safe to admit it is in the range starting at hundreds of kilowatts to megawatts.
According to thermodynamics, energy cannot be created or destroyed. Therefore, the heat removed from the object to be cooled must be delivered to another location that is willing to be heated up. Therefore, when cooling a hyperscale data center, the problem lies in dumping this enormous quantity of heat at a location that can accept it.
BSRIA, an organization involved in testing, instrumentation, research, and consultancy in the construction and building services, has recently conducted a market study. They offer a valuable insight into the cooling options and trends available to hyperscale data centers.
According to the BSRIA report summary, they have represented the feasibility and popularity of techniques versus data-center temperatures in a four-quadrant graph. Among the options shown are reducing dissipation by using modular DC power supplies and variable-speed drives, cooling techniques by using adiabatic evaporation to liquid cooling and allowing a rise in the temperature at the server-inlet. The graph includes growth potential against the investment level necessary for each approach – most popular is the adiabatic/evaporative cooling.
The adiabatic/evaporative process of cooling uses a natural phenomenon to regulate temperature. The cooler uses a large fan that draws in warm air through pads moistened with water, which evaporates. Huge quantities of heat are removed when water evaporates, chilling the air, which is then pushed out to the room. Temperature control is a simple matter of adjusting the airflow of the cooler.
For data centers and other facilities, the adiabatic/evaporative process has saved the industry millions of liters of water. Older cooling towers would pollute the water they used. Adiabatic cooling units also save greater than 40% in electricity consumption.