Energy-saving technology for data center servers

Energy-saving technology for data center servers

Every time you walk into the data center room, you can see that the vast racks are filled with servers. In a data center, the number of servers determines the service processing capability of the data center. In medium and large data centers, thousands of servers are used for calculation, ten servers are placed in one rack, and hundreds of racks are required for large data centers. Such a huge number of server devices run at the same time. Based on the computing power of 10 thousand servers, KW is required for servers. Therefore, if each server can save a little energy consumption, the energy consumption saved by 10 thousand servers is considerable. Therefore, many experts start from the server and work hard on the server to minimize the energy consumption of the server without reducing its running performance. The following describes the energy-saving technology of the server.

As we all know, the server has several major components: CPU, memory, disk, Nic, and motherboard components. CPU power consumption accounts for 1/3 of the total server power consumption. The stronger the CPU computing power, the larger the power consumption. memory is the most energy-consuming component except the CPU, the mainstream DDR3 and FBD memory have reached the order of 10 W, the higher the memory frequency, the higher the power consumption; the disk is also the main contribution of server power consumption, the current server hard disk specifications are very large, this naturally leads to an increase in the power consumption of the entire server. The power consumption of these main components increases, and the main board components, bus, and peripheral devices that work with them need to work at a higher frequency to make full use of their performance, this means that the overall server energy consumption increases. These are the most important components of servers and are also the largest energy-consuming users. servers can't do without any of these components, but also need to save energy on these components. As long as the energy consumption of these components is reduced, the energy consumption of peripheral devices can be reduced, thus reducing the overall energy consumption of servers.

Since the CPU is the most energy-consuming part of the server, let's talk about the CPU first. There are three main types of CPU energy-saving technologies: C1E energy-saving technology, that is, enhanced deep sleep technology. With C1E technology, the CPU can reduce the operating voltage and frequency doubling in idle and light load conditions. Intel's CPU also supports an EIST energy-saving technology. By enabling EIST, the CPU can control its frequency and voltage based on actual usage, thereby realizing power consumption control. AMD also launched "cool and quiet ". It is also an energy-saving technology that enables the processor to automatically reduce the voltage and frequency while idle. All these CPU energy-saving technologies allow the CPU to work on demand and reduce the working frequency when idle, so as to reduce power consumption.

Memory is the most energy-consuming part except the CPU, and the memory technology has been continuously improved over the past few years. Such memory has lower voltage, when there is no read/write operation, you can use the Reset command to stop all operations and switch to the least active status to save energy. New technologies are emerging in this field, such as the magnetic memory designed by the study team at the University of California, Los Angeles, with extremely low energy consumption, high density, fast reading and writing speeds, and no volatile data storage without power-on. Intel and Micron jointly launched the 3D Xpoint memory technology in July this year, which prompted another server giant, Hewlett-Packard, to work with SanDisk, to launch a new memory chip, the new technology is claimed to be 1000 times faster than traditional flash memory. Working hard on memory cabling density can also achieve energy-saving purposes. Generally, the memory is 50nm, and 50nm refers to the connection width of the semiconductor component substrate circuit to 50nm. Now there are 40 nm or even 30 nm, this means that more circuits can be accommodated in the same space.

A disk is the most important part of a server. Once an exception or power loss occurs on the server, data will not be lost as long as the disk is still there. However, if the disk is broken, data loss may occur. Disk energy-saving technologies mainly include disk sleep and disk speed reduction technologies. These seem to be based on the disk load (access volume) to control the disk to work in four states: full speed, idle, sleep, and powered off. Full speed means that the disk is running at high speed and data is being read and written. Idle is only a short idle period during the read/write period. If the disk does not work for a certain period of time, it enters the sleep state; if the disk does not work for a long time, it enters the power-off state. During the operation of the disk, the disk should be in the working state and automatically adjusted through software, so as to achieve energy-saving effect. Of course, too frequent switching of the working status will also affect the disk reliability. We also need to consider that the application business can tolerate the delay caused by switching from the low-power mode to the high-speed mode, therefore, the energy-saving technology adopting such dynamic adjustment should be considered comprehensively, and The introduction should be cautious. In addition to adjusting the operating status of the disk, you can also use a variety of speed disks, and use an SSD hard disk instead of a mechanical disk.

In addition to the above three components, Nic, fan, motherboard, display, indicator light and so on are all power-consuming components. For example, the monitor is not used for servers in the data center and is only used for debugging. Therefore, the monitor does not need to be considered. The fan in the server should also be considered. Many server fans are old-fashioned fixed-speed fans that run at the maximum speed. Currently, the server fan introduces the smart fan technology, by using a temperature sensor embedded on the server motherboard, the fan speed is controlled based on changes in ambient temperature, and the fan speed inside the server is properly adjusted. Although the network adapter occupies a low power consumption, you can adjust the power consumption by the busy degree of the network adapter link, and reduce the speed of the network adapter link based on the network adapter traffic to achieve energy saving.

There are many energy-saving areas inside the data center server. From these energy-saving technologies, it is not hard to see that they all follow the same design principle, that is, distribution on demand. Provide proper power consumption based on the actual running status of the component, so that no excessive power consumption is wasted. If every part of the server can save a little bit, it will not be able to accumulate, and then put it on the entire data center server, the amount of energy saved will be amazing, therefore, the research on server energy saving technology is an important part of the exploration of energy saving and emission reduction in data centers. By saving energy on server components, the whole server can be controlled in an energy-saving manner, and the data center energy-saving control can be achieved from the root.