On November 6, Xie Ming, vice president of Tencent cloud and general manager of Cloud Architecture Platform Department, disclosed the evolution and innovation of Tencent infrastructure for the first time at Tencent tech developer conference. According to him, behind the continuous development of wechat, QQ and other national businesses, the infrastructure and technical architecture including servers, networks, IDC, computing, storage, databases, etc. have been constantly evolving, and have entered a new stage of "born for cloud".
In terms of infrastructure construction, Tencent has more than 1 million servers in the whole network, which is the first company in China with more than 1 million servers in total, and the peak of Tencent network bandwidth has also exceeded 100t.
Xie said that in the cloud era, Tencent's infrastructure will continue to evolve to the direction of extreme flexibility and intelligence through self-research, customization and other means.
In terms of servers, Xie Ming said that Tencent cloud started the process of self-developed servers for the higher requirements of server performance, cost and security. Tencent's self research server, Xingxinghai, adopts the hardware tamper proof ability transmitted by the trusted chain, advanced thermosyphon heat dissipation technology, refined air duct design and fan control, realizing 30% TCO reduction, 50% load energy efficiency improvement, and 35% case comprehensive performance improvement.
Xie Ming mentioned that Tencent cloud's fourth generation data center T-block has realized the modular configuration and rapid construction of the whole data center in the way of building blocks. Compared with the traditional large-scale data center, the on-site construction cycle is shortened by half, and the natural cooling technology can reduce the pue to less than 1.2, which is the leading in the industry. With the T-block scheme, a park with 300000 servers can save 250 million kilowatt hours of electricity a year.
In terms of storage, Tencent developed its own distributed storage TFs for Q-zone photo album business scenario as early as 2006, which supports 100 million level picture uploads of qzone photo album every day. At present, Tencent cloud object storage cos has realized the system throughput of all nodes in the cluster, and the disk utilization rate has reached more than 90%. At the same time, the full autonomy of the system has also greatly reduced the labor input of operation and maintenance.
With the rapid development of cloud computing, cloud computing users put forward higher requirements for resource management granularity, resource scheduling and isolation capabilities, and the technical architecture of computing / storage decoupling. Xie Ming called this higher requirement "extreme elasticity".
Xie Ming believes that the ultimate flexibility is first of all that the performance and capacity should not only help big customers to bear the business peak, but also enable small customers not to waste a penny. It is necessary to divide computing, storage and network resources in a fine-grained way. Secondly, we need to improve the inventory management and scheduling capabilities and the corresponding product capabilities, so as to truly achieve the billing by usage. Finally, the fast delivery capability requires that all kinds of resources, such as computing and storage, can be fully decoupled to meet the requirements of rising and falling seconds, fast up and down.
Tencent cloud's basic products are also in the direction of ultimate flexibility. Among them, the third generation of general computing platform built by Tencent cloud without service function computing can provide better flexibility. In a real sense, the resource utilization rate is 100%, and the corresponding cost of users is reduced by more than 50%. It is worth mentioning that Tencent cloud uses function concurrent monitoring data real-time calculation to expand its capacity in advance, reducing the cold start rate of functions to less than 1 / 10000. The self-developed lightweight virtualization technology can start a virtual machine within 35ms, creating a record that the industry can start a virtual machine as soon as possible. This technology is also stably applied to 'small program & middot; cloud development'.
At the same time, Tencent cloud object storage cos also provides different billing methods in different regions. Users can choose the most cost-effective scheme according to the use scenarios. The database product, cynosdb, is based on the computational storage decoupling architecture, which supports large instances and achieves better flexibility. Single instance can reach 100 TB level at most, while storage sharing among multiple computing nodes can expand the number of computing nodes in the same instance to 15.
Xie also believes that in the future, cloud computing will combine intelligence to provide better services, including intelligent operation and intelligent services, which will better improve service quality and reduce service costs, and provide more users with VIP level cloud services. Taking database as an example, Tencent cloud provides online performance optimization services based on cdbtune, a database performance optimization system of reinforcement learning, even better than full-time DBAs.