Energy saving and carbon reduction, PUE is not omnipotent
Referring to the data center, there are two obvious labels at present. One is the base of the digital economy. As a new infrastructure, the data center is an essential key to the improvement of computing power and digital transformation; First, large carbon emitters. Statistics show that in 2021, China's data centers will use 216.6 billion kilowatt hours of electricity, accounting for 2.6% of the country's total power consumption, and carbon emissions will reach 122 million tons.
If no measures are taken, along with the expanding industrial scale, the carbon emissions of the data center will rise rapidly, which will eventually affect production and life. In this case, carbon reduction has become a matter that the data center must pay attention to and need to continue to promote, in order to achieve green and sustainable development. Among many carbon reduction measures, energy conservation and power reduction are the most concerned.
Carbon saving must be greatly energy-saving
The data center will generate carbon emissions in the process of construction and operation. Compared with the two, most of the energy consumption and carbon emissions are generated during operation, which is also the main part of this paper.
Taking the data center with 1000 cabinets and 2.5 kilowatt hours of single cabinet as an example, its annual power consumption during the operation period can reach 21.9 million kilowatt hours, and the operation cycle of the data center can be as long as 20-30 years. With the characteristics of uninterrupted operation, the data center needs long-term continuous power consumption, and most of the power is supplied by the grid. At present, thermal power (high emission fossil energy power generation) still plays an important role in the energy structure, This has also become the main source of carbon emissions in the data center. Therefore, reducing carbon emissions is actually a process of gradually reducing and offsetting thermal power.
From this point of view, the current relevant measures can be divided into three levels: in terms of energy consumption, use a variety of energy-saving means to reduce power consumption in all links and achieve the improvement of energy efficiency in all stages; In terms of energy supply, we will increase the proportion of new energy use, optimize our own energy use structure by equipping distributed new energy power generation equipment, green power direct supply and other ways, increase the electricity generated by new energy, and reduce the proportion of thermal power use; In terms of offsetting, it is to purchase green electricity, green certificate and other market behaviors for carbon absorption.
If you want to achieve low carbon, or even zero carbon, these three aspects are indispensable. In particular, energy use measures are indispensable for carbon reduction, and even the most important part at the current stage.
The reason is that compared with the change of energy supply structure and the marketization of carbon offset, the substantial reduction of energy consumption is a fast, effective, intuitive means that can significantly reduce carbon emissions in a relatively short period of time, and has the conditions and potential for further development in technology and market. At the same time, as the main body of energy use, the data center is more proactive in its practice and development in this regard, and the relevant energy-saving technologies and solutions are more innovative and can have better practical results. It is a reflection of enterprises' active social responsibility and a necessary measure to promote the industry as a whole to achieve practical carbon reduction.
Closely related to PUE, reducing energy consumption for cooling, power supply and distribution becomes the focus
The energy consumption of the data center includes IT equipment, refrigeration system, power supply and distribution system, lighting and others. Among them, reducing the energy and power consumption of refrigeration, power supply and distribution systems has become a major concern. This is mainly because both are related to PUE.
As an important indicator to measure the energy efficiency level of data centers, PUE has become the most popular word in the industry in recent years.
National and local policies have been issued to continuously promote the reduction of PUE value. In 2017, the "13th Five Year" Comprehensive Work Plan for Energy Conservation and Emission Reduction proposed that the PUE value of new large-scale cloud computing data centers should be less than 1.5; In 2019, the Guiding Opinions on Strengthening the Construction of Green Data Centers proposed that the PUE value of existing large and super large data centers should not be higher than 1.8; In 2021, the Three Year Action Plan for the Development of New Data Centers (2021-2023) proposed that by the end of 2023, the PUE of new large and above data centers will be reduced to below 1.3, and that of severe cold and cold regions will be reduced to below 1.25; At the end of 2021 and the beginning of 2022, the National Development and Reform Commission and other departments proposed that the average PUE of the eastern data center cluster should be less than 1.25, and that of the western data center cluster should be less than 1.2 in the reply of "calculating data from the east to the west".
It can be said that low PUE has also become a necessary condition for the data center to be able to construct, transform and operate. and PUE is the ratio of the total power consumption of the data center to the power consumption of IT equipment in the data center. To reduce PUE, you need to use all the power for IT equipment as much as possible to reduce the power consumption of other parts. Energy saving and power reduction measures in energy use are just in line with this characteristic.
Refrigeration is the largest energy consumption in addition to IT equipment. Taking the data center with PUE 1.5 as an example, its IT equipment consumption accounts for about 67%, the refrigeration system accounts for about 27%, the power supply and distribution system accounts for about 5%, and lighting and others account for about 1%. Reducing its energy consumption is the main way to reduce PUE, and it is also an important part of reducing power consumption in terms of energy consumption.
By optimizing the original cooling mode, enhancing the use of natural cooling sources, introducing advanced technology, etc., the data center can effectively improve its efficiency and reduce the use of electricity. For example, liquid cooling, which is discussed most recently, has a specific heat capacity that is far higher than that of air. The heat capacity per unit volume is more than 1000 times that of air cooling, with higher heat transfer efficiency. The power consumption of the system is about 30% - 50% less than that of the air cooling system, At present, there are three forms of spray type, cold plate type and immersion type, which are gradually applied to the data center.
In addition, the exploration and application of the data center in indirect evaporative cooling, full frequency conversion fluorine pump technology and other aspects have also effectively reduced the power consumption of the data center in cooling and improved the cooling efficiency. At the same time, In order to achieve the theoretical cooling effect, there are still many areas that need further optimization, such as perfect market system, mature service system, lower cost, more prominent safety assurance, etc. The development of refrigeration is still a long way to go.
In the third largest energy consuming part - power supply and distribution system, the data center has also made great progress, such as the application of high-voltage DC power supply system. Compared with traditional UPS, it has the characteristics of high system efficiency, convenient system expansion, high system stability, high maintainability, etc., which can effectively improve the power supply efficiency and play a good energy-saving effect.
When using HVDC, the newly built data center will usually use mains power+HVDC dual power supply after comprehensive consideration of cost and reliability. The power supply efficiency of this method is 98%, which eliminates the bottleneck of single point of failure of the system, improves the reliability of power supply, and provides AC/DC two-way power supply in each rack. There is no need for power conversion in the mains power, It can maximize system efficiency.
Through various measures of energy conservation and power reduction, part of the energy consumption outside the IT equipment of the data center is reduced, and PUE shows a continuous downward trend. According to statistics, compared with 2017, the average PUE value of China's ultra large data centers in use in 2019 decreased by 0.17.
Avoid the PUE only theory, and reduce the power consumption of IT equipment is a necessary link
There is no doubt about the importance of PUE, but if only PUE theory, it may have the opposite effect on reducing the power consumption of the data center.
This is mainly because PUE is one-sided For example, the application of water cooling storage can provide redundant emergency cooling for the machine room. By using valley electric power for cooling, it can effectively save electric energy, and involves the use of solar energy, water energy and other clean energy, which can improve the consumption of clean energy. If the cooling capacity is released to the environment and equipment for cooling during peak power consumption, it can also play the role of peak shaving and valley filling. However, in terms of PUE, if water cooling storage is used, the power consumption will increase, which may cause the increase of PUE.
This may also happen in terms of energy conservation of IT equipment. If the efficiency of IT equipment is improved through virtualization technology, the total energy consumption of IT equipment will be reduced, but the energy consumption of IT equipment will also be reduced, resulting in an increase in PUE.
However, IT equipment is the largest power source during the operation of the data center. Therefore, the data center wants to further save energy, reduce carbon emissions, and improve energy calculation and efficiency. Energy saving of IT equipment cannot be bypassed. Therefore, To reduce the power consumption of IT equipment, PUE cannot be judged only from PUE, or PUE should be comprehensively measured. In the process of promoting energy conservation and power reduction of IT equipment, certain standards should be set, and relevant factors should be subtracted as variables to obtain meaningful values.
What solutions or explorations does the data center have in terms of power reduction of IT equipment?
In terms of chips, DPU, which is more suitable for large-scale computing scenarios such as the data center, was born. It is a dedicated processor constructed with data as the center. It uses software defined technical routes to support a series of functions such as resource virtualization, storage, security, and computing in the infrastructure layer. DPU is required to break through lower latency and higher throughput in terms of core capabilities.
According to the test, by using NVIDIA ConnectX smart network card (network card with DPU function) to process network tasks, the CPU energy consumption of servers running on the full load network decreased from 190 watts to 145 watts, a decrease of 24%. With DPU, a large data center can reduce the power cost by nearly $2 million in three years.
In terms of storage devices, The basic core function of the data center is data storage, but the energy consumption of storage is also extremely huge. Data shows that the data storage system accounts for about 35% of the power consumption of the data center.
In view of this, in order to improve storage capacity and reduce energy consumption, the data center has upgraded and optimized the storage devices, and the representative product is flash memory. Statistics show that flash drives can save more than 65% energy than mechanical drives. The full flash data center has also been launched, with outstanding performance in energy saving and carbon reduction. For example, Industrial Bank implemented full flash replacement by deploying advanced flash memory devices, and the storage system realized more than 60% energy saving.
In terms of energy-saving IT equipment, High conversion efficiency power module, server fan energy-saving control, whole machine cabinet power supply, liquid cooling server, etc. become the research direction, which improves the efficient use of IT equipment and meets the development trend of high-density data center while reducing energy and power consumption; In terms of software system, Intelligent technology is used to move the workload in real time through optimization of architecture and algorithm to maximize the use of active devices, and put unnecessary devices in deep idle/low power state to improve the efficiency of use and reduce the demand for power.
In addition, Optimize the application and system architecture, For example, reducing unnecessary redundancy and backup, and improving program operation efficiency are also effective ways to promote energy conservation of IT equipment.
From a practical point of view, most of the above methods are still in a relatively early stage, including the R&D and application of chips, the innovation of intelligent software systems, etc., but their effects on energy conservation and power reduction have begun to show signs. At the same time, With the further reduction of PUE, the energy consumption of other equipment is greatly reduced, and the energy conservation of IT equipment is more meaningful, which will become an important way for the data center to improve energy efficiency.
Energy conservation and power reduction in the data center is an effective way to promote carbon reduction and a direct way to improve computing power and efficiency. In the actual process of promotion, it involves multiple industries such as liquid cooling, chips, storage, etc. It is a systematic project in which the data center is the demander and requires multi-party cooperation. At the same time, it also has its own complexity. The effect of relevant measures should be judged more comprehensively, not only by the current PUE, but also by the establishment of a more complete energy consumption evaluation system. At the same time, not only the operation, but also the construction of the data center should be included. The adoption of prefabricated modular and other more energy-saving construction methods is also an important measure for the data center to save energy and power.
