Cloud computing is undergoing new changes. On the basis of strong computing power, business scenarios are driving technological innovation and change. Tencent cloud officially released the third generation cloud server (CVM) matrix, and the latest computing, storage, network and heterogeneous computing examples were unveiled for the first time. So far, the cloud server (CVM) matrix provided by Tencent cloud contains 26 examples, covering 192 business scenarios such as e-commerce, video, games, finance, gene sequencing, intelligent voice, automobile, medical treatment and the Internet of things, enabling the acceleration of computing in the whole industry.
Cloud server is the most basic module in cloud computing services. According to the actual needs of different business scenarios, Tencent cloud server product matrix provides users with the most complete instance options in the industry. Users can build services based on different instances according to business needs.
In the general scenario, Tencent cloud launched three new instances of standard, computing and memory based on Intel’s new generation skylake CPU. Its computing performance was improved by 60% as a whole, and it provided an intranet bandwidth of up to 25gbps. The network receiving and contracting performance was twice that of the previous generation. Based on the instance of AMD epyc Series CPU, the instance configuration with the maximum 128 core specification is provided to effectively meet the calculation requirements of customers for a large number of CPU cores.
For the storage business scenario, Tencent cloud has launched two new storage optimization examples: big data model and IO intensive model. The big data model adopts throughput HDD local disk, and the sequential read-write bandwidth can reach up to 2.3gb/s, which is 4.7 times higher than that of ordinary local disk, which is suitable for big data services with high disk throughput requirements; The IO intensive model adopts the local nvme SSD hard disk, and the random IOPs can reach up to 100W IOPs. Compared with the SSD local disk, the random read-write IOPs is increased by 13.3 times, which is suitable for businesses requiring high IOPs performance.
For business scenarios requiring high network performance, Tencent cloud launched a new network optimization example. This example uses a new intelligent network card to offload the implementation of the virtual switch on the host to the intelligent network card. The packet forwarding is accelerated through hardware, so that the network forwarding performance is improved from the original 40W to 450W, and the network bandwidth is up to 25gbps. The network delay is greatly reduced, and the network delay close to the physical network card is realized. This can not only meet the needs of users for high-speed computing, but also enable users to get a performance experience comparable to physical machines in the virtualized network.
As the fastest-growing cloud service provider in FPGA cloud services in China, Tencent cloud will also launch two new generation FPGA computing examples. Among them, the fx3 instance based on Xilinx vu9p FPGA card supports up to 8 cards. Each card contains about 2.5 million logic units and 6800 digital signal processing engines. The peak computing performance is three times higher than that of the previous generation FX2 instance; Based on the fi3 instance of Intel stratix10 FPGA card, the single machine supports up to 8 cards. Each card contains about 2.8 million logic units, 5760 digital signal processing engines, and the single precision floating-point operation performance reaches 80tflops.
The network architecture of the data center is changing from 10 Gigabit to 25g, and the 25g network environment will become the mainstream of building the next generation data center network architecture. Tencent cloud’s third-generation server will fully launch a 25g network environment in the second half of this year. The intranet bandwidth will be expanded to 25gbps, the packet forwarding rate will be increased by more than 100%, and the delay will be reduced by about 30%. It can fully support business scenarios with high real-time requirements such as game business, video business and financial analysis. In addition, Tencent cloud has also launched 100g network construction in the laboratory environment.
The new generation KVM virtualization technology can support more hardware virtualization and provide richer functions. At the same time, it supports ultra-high speed thermal migration and can complete instance drift between different physical machines in 150 seconds, making Tencent cloud the first Internet cloud manufacturer in the industry to apply thermal migration on a large scale and stably.
With the continuous emergence of a series of emerging application scenarios such as gene sequencing, intelligent voice, unmanned driving and the Internet of things, the demand of enterprises for computing power is changing. Taking building stronger computing power and enabling the future of industrial intelligence as the starting point, Tencent cloud actively adopts the opening strategy of technological innovation and ecological construction, and cooperates with more partners to provide better enterprise computing services, so that more business scenarios can be landed in the cloud.
Taking FPGA as an example, for a long time, due to the need to use hardware description languages with low abstraction such as Verilog and VHDL in the development process of FPGA, it is difficult to develop. In addition, FPGA hardware has high cost, poor flexibility and large one-time procurement investment, which requires high cost and huge energy for FPGA customization and procurement. To this end, Tencent cloud cooperates with third-party partners to build FPGA cloud ecology, providing a series of excellent compression, encryption and decryption, image processing, video codec, genetic algorithm acceleration and other IP addresses. Users can easily integrate FPGA acceleration functions into their own applications by calling API.
On the one hand, this allows FPAG developers to enjoy the technical dividends of superior performance, convenient development and flexible billing of FPGA virtual machine, and migrate FPGA development from traditional hardware to the cloud. On the other hand, by establishing and expanding the third-party FPGA IP market, the third-party FPGA developers can provide their IP services in the FPGA IP market, while the third-party application developers can easily obtain IP resources from the FPGA IP market. The third-party application only needs to focus on the upper application and service development without paying attention to FPGA coding Debugging and other cumbersome details.