Energy Problems Faced by the Data Centers that Undergird the Internet
The Internet has become indispensable in our lives, and convenient services provided over the Internet improve year after year. These services are supported by data centers, which are facilities that operate services and applications provided over the Internet as well as company systems. In other words, when we use services on the Internet, we are using servers installed in data centers somewhere in the world.
As Internet use increases, data center operations also increase. Thus, power consumption at data centers continues to increase as well.
To solve energy problems at data centers, Fujitsu is participating in the Ministry of the Environment FY2016 Low Carbon Technology Research and Development Program. Recently, as part of the project, Fujitsu started demonstration experiments to achieve drastic data center de-carbonization.
Data Centers Consume 2% of the World's Energy
The amount of electricity used by data centers accounts for approximately 2% of total global energy demand, and this amount is increasing by 10% annually. Data center electricity costs have increased approximately eightfold over the past 10 years.
Internet services, particularly cloud services, have already become widespread in our lives and business, and they are expected to further expand in the future due to the use of big data, artificial intelligence (AI), the Internet of things (IoT), and so forth. In recent years, high performance computing (HPC), which refers to high-performance servers used for big data, AI, and the like has been rapidly growing. HPC offers high processing capacities but the electricity consumption and calorific value are also high. If the calorific value increases, the energy required for cooling also increases.
If Internet use continues to increase, data center electricity consumption will continue to increase more and more. So, we now need to seriously work on saving energy at data centers for the future.
Breaking the Barrier of PUE = 1.1 to Realize Zero-power-consumption Cooling Facilities
Large data centers have long worked to save energy. Still, small- and medium-sized data centers are said to lag behind in such efforts.
Consider data center power consumption from the viewpoint of power usage effectiveness (PUE), an indicator of energy efficiency. PUE indicates the ratio of ICT equipment power consumption to total data center power consumption. Besides ICT equipment such as servers and storage, data centers also contain cooling facilities and power outage support equipment that consume much electricity.
To operate systems, ICT equipment requires electricity. Therefore, how to control power consumption for non-ICT equipment is the key point for energy saving efforts at data centers. The ideal value, a PUE of 1.0, represents when the power consumption of the entire data center and the power consumption of ICT equipment are equal. In other words, PUE = 1.0 means that all electricity is used for ICT equipment, and no electricity is consumed by cooling systems or other devices. The larger the value of PUE, the lower the energy efficiency level.
In the 2000s, PUE values generally ranged from 1.3 to 2.0. However, nowadays, almost all large data centers have achieved a PUE of 1.1 and started efforts to achieve a PUE of 1.0. Nevertheless, many small- and medium-sized data centers still have a PUE of around 2.0.
According to the EPRI *, 0.7% of data centers worldwide are classified as large, while the remaining 99.3% are small- or medium-sized.
It is not easy for non-large data centers to break the barrier of PUE = 1.1 because they cannot develop energy-saving technology easily on their own. Therefore, this project aims to develop highly versatile energy-saving technology so that small- and medium-sized data centers can achieve PUE = 1.0.
*: Electric Power Research Institute
The Best Techniques for Cooling High-heat-generating, High-performance Servers and Low-heat-generating Storage
Though over the years various energy-saving measures for data centers have been examined, in this demonstration experiment, Fujitsu focused on categorizing cooling methods by the heat generation level of ICT equipment. Equipment at data centers can be classified into one of three types according to the level of heat it generates.
The first type is HPC, which has led to a rapid increase in the number of high-heat-generating servers in recent years. HPC servers offer high performance and consequently have high power consumption. Because of their high calorific value, they need strong cooling. At present, many types of equipment consume electricity within the range of 9 to 15 kW. In the future, the number of pieces of equipment that consume more than 16 kW and thus cannot be cooled by conventional air-cooling methods will increase.
The second type is servers with a medium-level calorific value that consume approximately 5 to 8 kW of electricity. The third type is storage with a low-level calorific value. Hard disk drives (HDDs) for recording data consume 4 kW or less.
In today's data centers, these three types of ICT equipment are operated on the same rack. When three types of equipment with different calorific values are mixed, the cooling capacity must be large enough to cool the HPC servers, which generate the most intense heat, and thus extra cooling power is required. This means equipment that generates low heat is cooled more than necessary.
To eliminate such waste, in demonstration experiments Fujitsu decided to place the three types of ICT equipment in different zones and to adopt the optimal cooling method based on the ICT equipment's calorific value.
Three Technologies to Efficiently Cool High-heat-generating Servers
Cooling system applicable to high heat-generation density
Of the three types of cooling systems, Fujitsu is in charge of a demonstration experiment on a high heat-generation density cooling system for high-heat-generating servers. In this method, the ICT equipment is completely submerged in a fluorochemical inert fluid (Fluorinert), which provides high heat-transfer efficiency and insulation as a coolant. Fluorinert is a clear, colorless liquid with low viscosity and high heat transfer efficiency (approximately 10 times as high as air cooling), and it is very stable thermally and chemically. Even if HPC servers are placed in a rack without space between them, sufficient cooling performance can be secured.
In addition, in this demonstration experiment, Fujitsu has taken on the challenge of adopting a new technology to cool servers by using the power of natural convection currents to circulate Fluorinert. This liquid immersion cooling technology using natural convection currents is unique to Fujitsu.
Various types of Fluorinert offer different heat efficiencies. As the cooling efficiency decreases if convection currents cause swirls, difficult control technology is required for natural convection. In this demonstration experiment, Fujitsu has attempted to select appropriate types of Fluorinert based on calorific values and to further increase precision through simulation.
Cooling system for medium heat-generation density
In a demonstration experiment on a cooling system for medium heat-generation density overseen by NTT East, an oil tank storing Fluorinert was placed on racks, and drops of Fluorinert fell like a shower on the substrate of medium heat-generation density servers. Fluorinert is designed to drip naturally by the force of gravity, so no electricity is used to make it drip. Moreover, even if a power outage occurs, the servers can be cooled until they shut down so long as the Fluorinert stored in the oil tank continues to drip.
Cooling system for low heat-generation density
In a cooling system for low heat-generation density overseen by Takasago Thermal Engineering, servers are cooled only by fans built into ICT equipment and natural convection currents, together with air conditioning. In this mechanism, cool air is sent from the front of the racks in a case; the air is circulated only by fans installed in the servers and the power of convection currents.
This project aims to achieve drastic de-carbonization of data centers by properly employing these three types of cooling systems and achieving PUE = 1.0 by reducing energy consumption as much as possible.
Automatic Cooling Adjustment Using AI
In the future, we will bring into view improved cooling efficiency using AI. As various factors, such as temperature and humidity, affect cooling performance, AI-managed automatic cooling adjustment is appropriate. To this end, we are also planning to accumulate operation know-how through this demonstration experiment.
To create a better society and environment, we must work on energy-saving measures for data centers now. Fujitsu will continue to strive to develop new cooling methods into the future, aiming to realize energy-saving data centers to achieve PUE = 1.0.