According to Synergy Research Group, the U.S. accounts for 49 percent of data center capacity in terms of critical IT workloads, China is the second largest data center capacity site after the U.S., accounting for 15 percent of the total, and the remaining capacity is distributed in other countries in the Asia-Pacific region (13 percent), Europe, middle east and Africa (19 percent), and Canada/Latin America (4 percent).
According to the Data Center Committee (CDCC), in 2019, the number of online running servers and data center electricity consumption in China were 12 million and 81 billion kWh, respectively, and the United States was 17 million units and 115 billion kWh.
Combining the above two sets of data, we can come to two conclusions, the average capacity of data centers in China is lower than that of the United States, but the power consumption per capacity is higher than that of the United States.
Figure | Technical Director of CommScope North Asia Wu Jian
In fact, data centers have always been a high-energy-consuming industry, with annual electricity bills of hundreds of millions of yuan, often referred to as electric tigers. In this regard, Wu Jian, technical director of CommScope North Asia, said at the CDCC in 2021, "The local energy consumption of the data center itself is showing an increasing trend, and the physical density is also increasing, many years ago, the power distribution demand of an ordinary rack was about 3-4kW, and now it needs 8-10 kW." To some extent, most of the costs in the operation and maintenance of data centers are to pay electricity bills, and the cost of electricity may account for 60%-70% of the operating costs of a data center. Therefore, as the energy consumption of the data center itself continues to increase, the dependence on energy will become more and more serious, and energy conservation and emission reduction will be put on the main agenda. ”
Regarding how to save energy and reduce emissions, at the specific operational level, in the past few years, the PUE has been reduced from the perspective of power distribution optimization and refrigeration optimization, which refers to the ratio of energy supply by infrastructure to IT equipment and energy consumption of IT equipment itself, and the smaller the ratio, the better. But in the past, we have reduced the PUE value from 1.5 to 1.35 and then to 1.3, which has entered a bottleneck period. So standing in today's dimension, jump out of the inherent thinking, the following will share with you about the two new dimensions of data center carbon reduction.
Two new dimensions of data center carbon reduction
No longer obsessed with PUE, start with the IT equipment itself to reduce energy consumption
First of all, we must understand a truth, our IT equipment needs some auxiliary equipment, such as cooling fans, etc., these auxiliary equipment is also to consume power, so the overall power consumption divided by the IT equipment itself The value must be greater than 1, even if it is completely natural cold, there will be other power supply. Based on the existing technology, 1.3 has dropped to a very low level, so we may wish to start with the "1" itself to reduce the power consumption of the IT equipment itself.
Wu Jian explained: "IT equipment currently has utilization problems in China's data center industry, such as 1,000 racks in a computer room, 10,000 servers deployed, but are they all in use? There is 1/3 of the area vacant in the computer room, and 1/3 of the server is vacant in the rack, is this empty consumption? In addition, green energy saving is also related to the connector, can we reduce the power consumption of the interface? Therefore, "1" itself is problematic, that is, the asset utilization rate of IT itself is not high enough. ”
Extend the lifecycle of domestic data centers
Figure | The life cycle of the cabling system
In addition to reducing the power consumption of IT equipment itself, we are digging a step deeper, can we extend the service life of the data center? As we all know, in the process of building data centers, a lot of industrial raw materials are needed, such as steel, plastics, chips, refrigerants, etc. The production of these raw materials will bring carbon emissions, which is a chain reaction, so extending the service life of data centers is itself in energy conservation and emission reduction.
When it comes to the life cycle of data centers, the life cycle of many data centers in China is too short, and it will be renovated in three or five years. This means that all IT assets and infrastructure will be scrapped and then a larger, more efficient, higher-rate data center will be rebuilt. In stark contrast, the design concept and materials used by large foreign Internet companies to build data centers now need to be considered for fifteen, twenty, or even twenty-five years, and this is also the case. Therefore, how to extend the life cycle of domestic data centers will become the focus of energy conservation and emission reduction in data centers.
Reduce data center power consumption with optimized network architecture and lower power connectors
"From an industry point of view, what CommScope can do is to improve the effectiveness of the data center, the data center is nothing more than the interconnection of IT equipment, and the interconnection of IT equipment involves network architecture issues." When it comes to network architecture, the trend is cloud architecture, using Leaf-Spine architecture or Fabric architecture, which can improve the performance of computing switching and storage inside the data center, reduce latency, and improve reliability. In addition, the development of high-speed Ethernet can further improve the network efficiency of data centers. Wu Jian said.
When we focus on connectivity, how can we reduce the power consumption of the interface, such as the use of more cost and lower power multimode fiber, thereby reducing the power consumption of the IT equipment itself.
As we all know, single-mode fiber needs a laser light source, multi-mode light uses a VCSEL light source, and the energy consumption of the VCSEL light source itself is much lower than that of the laser light source. From the perspective of light source wavelength, multimode is 850 nm, single mode is more than 1310 nm, most of which is between 1550-1650 nm, so single mode needs a longer wavelength, supports a longer distance, and requires higher power. From the perspective of wiring, there is not much difference between the manufacturing costs and energy consumption of the two, but from the perspective of equipment transceivers, under the same number of port applications, the energy consumption of single-mode devices is about two to three times that of multimode devices.
Figure | The evolution speed of the server
However, although multimode fiber is more energy efficient than singlemode, multimode has the limitation that the connection distance is too short. With the evolution of network applications from 40G to 400G, the transmission distance of multimode fiber is slowly reduced to less than one hundred meters, but if even one hundred meters cannot be supported, in larger data center applications, if the distance from the switch to the server is more than 100 meters, multi-mode will have problems, and it must be replaced by single mode, so how to talk about energy saving?
"So, over the past 3-4 years, CommScope has led the OM5 standard, which is a higher-grade multimode fiber. According to the development history of multimode fiber, we can arrange its standards: OM1, OM2, OM3, OM4, OM5. OM5 is a technical standard based on the evolution of multimode technology, which can further increase the rate, further extend the transmission distance and multimode vitality", Wu Jian explained, "so the emergence of OM5 is a multi-mode without rushing to upgrade to single mode, which is also an energy-saving idea." ”
Adopt an intelligent management platform to avoid blind upgrades and transformations
"CommScope has a product called imVision® Intelligent Cabling System, which is abbreviated as Automated Infrastructure Management (AIM), which is an intelligent cabling management platform. Through the platform, you can understand how many ports are in use in the entire IT asset connection, how many ports are idle, what is the asset utilization, and the properties of each port, such as what is the application of the connection peer, how far the link distance is, how high the rate, these can be managed, recorded, and reported in the system, so that users have a clear understanding of the wiring resources including IT assets and connections. From Wu Jian's statement, we can make it clear that understanding these parameters means not blindly upgrading and transforming.
Maybe the current capacity is enough for the next three to five or even ten years of applications, then there is no need to expand, which can further improve asset utilization. For example, the original port of 1000, now used 500, the next decade may only use 800 ports, there is no need to redeploy new resources. If you don't understand asset utilization, you may have to re-acquire, deploy racks, switches, cabling systems, and so on because of new business.
Figure | Intelligent infrastructure management
In addition, the intelligent management platform can also manage the power over Ethernet (PoE) application, which treats the data cable as a power supply cable. The camera or IP phone itself must both power supply and communication, and PoE can realize the combination of power supply and communication, but this is equivalent to giving the switch an extra task, the original switch is only used to send and receive signals, not for power supply. Now that the switch is used for power supply, its power consumption can also be managed through the platform.
How to achieve this real-time dynamic monitoring from the operational level? In fact, the data collection of the wiring industry is not the same as other industries (such as face recognition, intelligent camera), the traditional connection line is passive devices, and the wiring system is to establish a connection between the two active devices A and B, so we need to add sensors between these connections and connectors, and then electronically make the patch panel electronic, which is the concept of IoT.
Wu Jian said, "The energy consumption of the optical fiber transceiver is basically fixed, for example, after starting, it is providing 3-5 watts of power stably, so it can basically calculate how much its total energy consumption is." Copper cables are more complex, especially in smart buildings, and are more complex if there are PoE applications. ”
There is a particularly interesting phenomenon in Ethernet, such as plugging an IP device into a switch, first negotiating two things.
First they negotiate a rate between them. If one is a 100-megabit device and the other is a gigabit device, it ends up negotiating a 100-megabit device. If one is gigabit and the other is also gigabit, it may end up being negotiated as a gigabit. Negotiation means that it is variable, you don't know what rate it is, which can be seen from a switch management perspective, but it is variable.
Another consultation was the provision of energy. Ordinary Ethernet terminals provide data signals and do not provide energy. But if it is a POE camera, it may be negotiated, this camera to it to apply for 30W of power, whether the switch can provide depends on its capabilities, may only provide 15W instead of 30W; but if there is a capacity of 60W, it can provide 30W. To get 60W of power, the switch has to provide 60W. The process of providing energy consumption in this way is dynamic, and dynamic energy consumption needs to be monitored, which can be obtained from the intelligent platform to know what the total energy consumption is.
Therefore, this platform-based intelligent management can strengthen the optimization of equipment energy consumption and wiring design layout, so as to achieve the ultimate goal of energy conservation and emission reduction.
Write at the end
Regarding when the communication network of the data center will be upgraded to 800G, Wu Jian told the author, "Just like home decoration, water and electricity must be before the buyer of electricity." As an infrastructure, wiring is more forward-looking than network equipment, and 400G and 800G technologies have actually been ready in the past three or five years in terms of wiring. The sign of preparation is high-performance optical fiber and high density connectors, MPO connectors have appeared very early, such as 16-core, 24-core standards, which is actually for 400G, 800G service, from this point of view the wiring is ready. ”
"But whether the industry has reached a mature period does not depend on wiring, but on the maturity of the optical chip and the optical chip industry." The sign of this industrial maturity is two important indicators, one is cost reduction, and the other is energy consumption optimization. Despite their efficiency, high-speed transceivers consume more energy than low-speed transceivers. For example, the rate has increased by four times, and the energy consumption has only doubled, which looks good, but the user cannot accept it. Therefore, to solve the cost and energy consumption problems of photoelectric chips in the industry, of course, it also includes heat dissipation problems. Once the above two basic problems are solved, the rate of commercialization of this industry will certainly increase. ”