Lenovo thinkserver builds a new green data center for Baidu
recently, Baidu, the world's second largest independent search engine and the largest Chinese search engine, has established a new green data center. The newly built data center fully meets the background computing needs brought by Baidu's increasingly large number of visits and searches, and also enables Baidu to have a greener, energy-saving and environmental friendly IT infrastructure, laying a good foundation for Baidu's healthy and sustainable development
the new data center is composed of 1617 thinkserver rd530 two-way rack servers newly launched by Lenovo. The low-energy and high availability Infrastructure Solution customized by Lenovo for Baidu helps Baidu easily and stably cope with the huge traffic leakage. The experiment is to inject 1/5 volume of clean water into the plastic bag and the operation and maintenance pressure brought by the search volume, rich and diverse Internet service businesses, and effectively optimize energy efficiency at the same time, It has reduced the cost of ownership and operation, and effectively supported the expansion and improvement of Baidu's business
project background
Baidu is the world's second largest independent search engine and the largest Chinese search engine. Since its inception, Baidu has made it its mission to make it the most convenient for people to obtain information and find what they want. Since its inception, the company has adhered to the user-oriented concept, constantly adhered to technological innovation, and is committed to providing users with simple and reliable internet search products and services, including: functional search based on network search, community search based on post bar, and vertical search for various regions and industries, MP3 search, as well as portal channels, Im, etc., fully cover all the search needs of the Chinese Internet world, and also expand navigation services, community services, games and entertainment, software tools and other businesses. According to authoritative third-party data, Baidu's search share in China is more than 80%, and the number of search queries processed every day in the world has exceeded 5billion, far exceeding Google
demand analysis
Baidu's main business is search business. At present, the purchase server is mainly divided into two types: storage server and balanced server. For the search business, the storage server is mainly used to store all kinds of data (including data obtained by crawlers, retrieval databases, logs, etc.). It uses a 12 Bay low-end Intel server, adopts a distributed file system (file segmentation, distributed storage) and three copy technology (each piece of data is stored on four different hard disks, and these four hard disks belong to different servers), so it can meet the performance and reliability requirements by using the most cost-effective 3.5-inch SATA hard disk
the balanced server is mainly used for crawling, analyzing, indexer and other roles. It generally requires a large amount of memory and can achieve high-speed access speed. In order to improve the indexing speed and reduce the delay, it adopts the parallel computing system to realize the search and retrieval, and also configures the SSD hard disk in the server as the cache of hot data, and cooperates with Baidu's own software algorithm to realize the performance improvement (the random reading performance can be improved several times or even dozens of times)
with the continuous expansion of the purchase scale of user servers, Baidu has obtained the grade identification authorization from the American authority in 13 years. The total number of servers will exceed 200000, and users will face great pressure and challenges in terms of power consumption and operation and maintenance:
imagine that the power consumption of each server is 1 watt higher, and such a number of servers will be at least 200000 watts higher, It is conceivable that the electricity cost spent for more than a year (calculated at 0.8 yuan per kilowatt hour, the saved cost is 200000 1000 x 0.8 yuan x 24 hours x 365 days =1.4016 million yuan). On the other hand, in the computer room rented by users, the number of servers placed in each cabinet is mainly limited by current rather than bandwidth. At present, the current limit of each cabinet in the telecommunications room is about 12a. Based on the maximum power consumption of 250 watts when each server is running, 220V x 12a 250W =10.56 servers can be placed at most. For the risk of overcurrent, only 10 servers can be placed in one cabinet; If each server can save 10W of power consumption, then 220V x 12a 240W =11 servers can be placed in each cabinet, and one more server can be placed than before. If 10000 servers are placed, due to the reduction of power consumption, 1000 cabinets were used to place these servers, but now only 910 cabinets are needed. In this way, the occupied space of 90 cabinets is saved in the rented machine room, which saves a lot of costs for users
at present, baidu 100000 servers are bound to face the following problems in operation and maintenance management:
system online deployment
fault accurate alarm, which involves specific issues such as each memory, each hard disk, each fan, each power module and other
software and firmware batch update and modification, such as BIOS, BMC, raid card and other firmware update and maintenance
control the failure rate, Reducing MTTR time
each of the above links is a great challenge. Customers hope that the purchased servers can integrate internal and external resource management and fault diagnosis, which greatly improves efficiency and saves human costs at the same time; In addition, MTTR (mean time to repair) can also be reduced to a lower level; The redundancy and hot plug requirements of main components can also greatly improve reliability and maintainability
scheme design
Lenovo's new generation thinkserver server fully considered the user's application needs at the beginning of development, and made a lot of efforts in energy consumption optimization:
efficient power supply: it adopts the gold power supply with the highest standard in the industry, achieving 94% power conversion efficiency; Support dynamic load regulation: rd530 power supply can dynamically adjust the load of two power modules, evenly distribute the load of each power module, and ensure the conversion efficiency of power supply
wide body chassis: and the output air pressure of the experimental machine is in direct proportion to the driving air pressure. The chassis design is 6mm wider than the industry standard chassis, increasing the cross-sectional area of air flow, reducing wind speed, reducing fan speed, and reducing power consumption
hierarchical heat dissipation: give priority to heat dissipation of components with high calorific value such as processors and memory. Compared with the traditional four-layer heat dissipation design, it can reduce the resistance of high airflow and preheat processors and memory, and improve the heat dissipation efficiency of the system by more than 15%, so as to achieve the purpose of energy saving
star sensing technology: more than 50 sensors are densely distributed in the whole server system to accurately monitor the system status such as temperature, voltage and power consumption, so as to provide accurate data for the system fan speed regulation. Combined with the fully automatic speed regulation strategy, it can effectively save the power consumption brought by the fan under the condition of ensuring heat dissipation
thinkserver has also made a number of optimizations in operation and maintenance and management.
General Technology of main components improves the integrity and consistency of family products, reduces the difficulty of product maintenance, protects users' investment, makes hair look comfortable after use, and naturally brings convenience for users to manage the data center
fault self diagnosis module, which can quickly locate or warn the fault location, including CPU/memory/fan/power supply, etc; Rd530 adopts knife edge guide rail, including the screw free design of the upper shelf and the internal operation of the server
bmc adopts the general IPMI v2.0 interface, which is convenient for customers to develop centralized monitoring and management software; Easymanage management software is attached with the machine. It can be used across platforms by controlling the interface through the web. It can automatically identify the topology of devices in the network, realize the management of thinkserver products, and support the monitoring and management of physical machines and virtual machines
smart grid, an advanced energy-saving management software, helps system administrators effectively monitor energy consumption and equipment utilization, and can set strategies to reasonably reduce system power consumption
in addition to the above factors considered at the beginning of the design, Lenovo has also cooperated with customers in a number of software optimization work throughout the progress of the project to reduce power consumption and improve operation and maintenance management, mainly in the following aspects:
BIOS realizes the function that the CPU can work at the highest frequency of turbo in the idle state of the system, So as to reduce the delay in providing search services
bios realizes the function of modifying BIOS options under Linux system and improves operation and maintenance efficiency
bmc realizes NCSI function (BMC share NIC function can be flexibly adjusted to meet the operation and maintenance requirements of different computer room network environments)
bmc optimizes fan speed regulation strategy and further reduces power consumption
User revenue
Baidu adopts thinkserver server with the legendary gene of think and innovative design, Combined with easymanage, the centralized management platform of Lenovo server, a highly available, reliable and high-performance infrastructure has been established. The excellent energy-saving design of thinkserver not only improves the density of servers placed in cabinets and reduces the total cost of ownership, but also helps Baidu build a green and energy-saving data center, laying a solid foundation for the further healthy and stable development of Baidu's future business
Baidu believes that the new data center based on Lenovo thinkserver server fully meets Baidu's huge data throughput and computing needs, and establishes a stable and reliable IT infrastructure support for a more rapid, smooth and efficient search experience. This time, thinkserver not only supported the stable and healthy development of Baidu's businesses in all aspects with a stable, safe, efficient and easy to manage infrastructure platform, but also built a high-density, low-energy Green Data Center for Baidu with excellent energy-saving design, which not only saved a lot of costs for Baidu, but also established a green evergreen foundation for Baidu with IT infrastructure
LINK
Copyright © 2011 JIN SHI