Home > News content

Jinshan Video Cloud AI Edge Computing double acceleration, pilot 5G Ultra High definition era

via:博客园     time:2019/12/20 14:56:11     readed:561

With the arrival of the 5G era, the traffic flow in all industries is increasing, the video industry is particularly outstanding, and the whole Internet flow has been exceeded

However, with the continuous improvement of people's requirements for visual enjoyment, as well as the higher requirements for high-quality video in online education, live game, e-commerce, telemedicine, VR / AR and other industries, cloud service providers, as the underlying technical support, will also face new challenges.

In the new market competition environment, the change of the technology upgrading and business model of the cloud service provider becomes the 5G application

5g incoming video cloud opens a new voyage

At the meeting, Zongjie shared the keynote speech of "5g era video cloud opens a new voyage". Zongjie said that the current Internet video industry mainly embodies the four characteristics of the transmission network foundation, consumer entertainment oriented, hardware upgrade and rapid scale development. China's 5g has been officially commercialized. According to the white paper on 5g economic and social impact of the Chinese Academy of communications, the direct and indirect output driven by China's 5g will reach 6.3 trillion yuan and 10.6 trillion yuan respectively by 2030.


Zongjie, general manager of Jinshan cloud CDN and video cloud Product Center

5g's super high speed, super connection and ultra-low delay will promote the rapid development of video applications, mainly in the two aspects of entertainment consumption upgrading and industrial application popularization. The upgrading of entertainment consumption mainly includes family wide screen, VR cinema and cloud games; the popularization of industrial application involves online education, telemedicine, VR test drive and other industries requiring HD video.

In order to adapt to the development characteristics of 5g era, Jinshan video cloud has clearly defined the development direction as upward AI and downward marginalization. According to Zongjie, up AI is mainly to help users improve image quality and reduce costs. Jinshan cloud AI product series mainly includes Jizhi HD, KIE, kav1, image quality evaluation, etc.; down edge involves three aspects: larger capacity, deep edge and open edge computing.

For video applications such as HD video and VR, Jinshan Cloud Cloud Distribution Node is designed with special large capacity. The system architecture level adopts the application of the new generation Cache architecture Dorado, multi-level caching technology and multi-threading technology to improve the performance of concurrency and stability by more than 50%;At the hardware device level, the third generation DMA technology(SSDMemCacheAIO)is adopted, and the special optimization for compiler, storage and network is carried out, and the I/O performance is improved by 400%;At the level of transmission protocol, Jinshan Cloud TCP network is improved by 30%through kernel optimization, and the support of QUIC protocol makes the download speed of weak network obviously.

At the same time, Jinshan cloud realizes the edge calculation of internal services through rapid deployment, gray-scale isolation, resource reuse and accurate statistics, and opens the edge calculation ability to the outside world, providing multiple product forms such as edge host, edge container and edge PAAS.

In addition, Jinshan video cloud is also actively expanding overseas business, and overseas self built nodes cover all important countries along the route. At the same time, jinshanyun cooperates with well-known foreign manufacturers to help domestic enterprises go overseas.

Edge computing kenc powerful computing force support and time delay control

Then, Sven gave a speech with the theme of "Introduction to cloud edge computing in Jinshan". In the 5g era, with the continuous rise of industrial gateways, instant messaging, cloud games, AR / VR and other applications, the requirements for bandwidth and delay are getting higher and higher, and edge computing is on the rise. In general, edge computing nodes can be deployed in Metro, base station and home, each location has its own advantages. The metropolitan area deployment can cover 200000 people, with a delay of less than 10ms, strong computing power and standardized environment, and the largest capacity and bandwidth scale. Therefore, Jinshan cloud chooses to calculate the kenc on the edge of the metropolitan area layout.


Sven, general manager of Jinshan cloud edge computing product line

Kenc is a set of container cloud platform running on the edge node of Jinshan cloud CDN. It supports the user-defined container image running on the edge. Relying on the main container cloud platform of CDN network, it provides automatic control node container cloud for large customers and automatic scheduling container cloud for small and medium customers. Through the principle of nearby computing, kenc efficiently distributes computing power to the whole area, so as to provide high-performance and low delay edge computing services. It combines the super processing ability of the cloud with the smooth operation experience of local devices, improves the user's experience, and helps customers explore more feasible business models.

At present, Jinshanyun has 30 large core data centers, 1000 PB data storage, 1000 video CDN distribution nodes and 100T network bandwidth reserve. Jinshanyun CDN edge hosting constructs the underlying technology of edge computing, and provides container mirror, template customization, application distribution and service opening capabilities. At the PaaS level of edge computing, the speed and availability of API, global data synchronization and intelligent node scheduling are solved through the interconnection of edge nodes to solve the problem of edge access to massive KV data.

The cloud edge cloud computing, KENC, provides IaaS PaaS resources, solution consulting, and collaborative development to provide customized services to users, Sven said. At present, the cloud edge of Kingsoft is calculated in the large data uploading and API acceleration scene, the uploading speed of the file is increased by 40%, the uploading success rate is increased to 99.5%, and the success rate of the API service inquiry request is increased to 99.9%; in the scene of the same city live broadcast, the friends and the relatives of the same city are allowed to watch the nearby drawing stream, greatly saving the bandwidth of the backbone network and improving the viewing quality; in the cloud game scene, the problems of large space, insufficient client configuration, cross-platform of the game and the like of the game client are solved; and in the two practical scenarios of the IoT, the industrial IoT equipment is extremely easy to be arranged; The household IoT equipment has insufficient hardware capacity due to price subsidies. The two types of IoT devices can move the computing logic to the edge node, the edge node is powerful, the delay is controllable, and the real-time update is achieved.

Ji Zhi HD AV1 Cloud Transcoding leads the 5G Ultra HD era

Then, Cai Yuan shared the keynote speech of "smart HD AI capability to build smart cloud transcoding". Caiyuan said that Jizhi HD is an intelligent video solution based on AI technology and traditional video coding technology. It integrates a variety of visual AI analysis algorithms, analyzes and identifies the video content through deep learning network, optimizes subjective experience and adaptive decision-making coding parameters, and pursues the ultimate bit rate compression while ensuring high-definition human visual experience. For nearly a thousand test videos, the average bandwidth saving is 61.34%, and the transcoding quality exceeds the original image quality through subjective evaluation.


Cai Yuan, senior technical director of jinshanyun

Using advanced antagonistic neural network technology to generate high-definition images, KIE can achieve 8 times to 16 times super score, and the human eye perception after super score has no sense of violation. At the same time, Jinshanyun officially provides AI AV1 intelligent cloud transcoding service, the compression ratio is as high as 80.57%, the pursuit of extreme video compression.

In the practical application case, the AI series products are widely accepted by the user. In June 2019, the image compression service with intelligent high-definition extension is provided for the Xiaomi products, and the image transmission bandwidth of more than 50% can be saved without changing the JPG format; in November, The high-definition H.265 cloud transcoding service is provided for the Xiaomi Mall, and the bandwidth is saved by 56.3% under the condition that the image quality is obviously enhanced; in November, the on-demand and intelligent high-definition H.265 cloud transcoding service is provided for a well-known short video client, and the code rate saved on the basis of restoring the image quality of the UGC is up to 40%; in December, Help an online education client to integrate the streaming service of the live PC end encoder, and the bandwidth saving is up to 41% in the case of the unchanged subjective image quality.

In terms of the latest coding and decoding technology, fan Hongfei said in the keynote speech "Jinshan cloud video coding technology leads 5g HD era", Jinshan cloud has launched kav1 commercial encoder based on the fourth generation coding standard AV1. Compared with h.265, more than 70 new toolsets are added to AV1 encoder, and the support of mobile phone and Web terminal is more complete, which also accelerates the process of commercial landing. It is expected that AV1 will begin to appear commercial landing in 2020. Jinshan cloud's kav1 encoder gathers more than 40 original algorithms of six processing modules, namely preprocessing, code control, prediction, transformation, entropy coding and post-processing, to create a professional commercial encoder for various scenarios.


Fan Hongfei, algorithm architect of jinshanyun

Kav1 customized and optimized the seven scenes and six evaluation indexes, combined with Chi HD video classification and quality classification, and automated decision-making coding optimization parameters. In the game, film, show, variety show, party, monitoring, education and other scenes, neutrality can be greatly improved. Fan Hongfei said that in 2020 kav1 is expected to increase to 30 times faster than svt-av1 in slow speed. AV1 combined with smart HD is expected to achieve 85% code rate savings when subjective video is superior to original video.

Zero fault jinshanyun escorts major events

In the end, Chen haibiao shared several major events of major customers such as CCTV, the escort of Jinshan video cloud.


Chen haibiao, deputy general manager of Jinshan cloud CDN and video cloud Product Center

One is the case of flexible service. On the evening of August 18,

Second, the case of high concurrent processing capacity, jinshanyun participated in CCTV Spring Festival Gala, National Day military parade and World Cup re insurance activities. In the National Day military review project, the peak bandwidth of single event history is 30 TBPs, and the maximum number of concurrent users reaches 12 million. Through the new generation of self-developed cache system and hierarchical cache technology, Jinshan cloud has explored the hardware potential and improved the service ability by 50%; at the same time, with edge data statistics, it has expanded the data processing ability horizontally to ensure real-time data feedback.

Through the integration of monitoring, alarm, operation, operation and maintenance, big data, AI analysis, and automation strategy formulation, an automatic full link system guarantee is formed to effectively cope with the peak traffic and ensure the smooth operation of the National Day parade live broadcast. From the beginning of data collection, to the establishment of the front field security and rear duty arrangement of the heavy security virtual team, to the process sorting of key emergency plans, to the routine, reporting and fault emergency drills, to the peak processing during the National Day activities, Jinshan cloud has a mature heavy security mechanism plan and service system. In addition to the gold service guarantee for major events, jinshanyun is more able to provide daily services for customers.

At present, Kingsoft video cloud has served all customers of video industry TOP20, more than 80% of well-known live and short video APP, and has been widely recognized by the industry.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments