A New Paradigm for Video Cloud Native Services

On November 5, 2022, the special session of "Relying on Heaven to Open a New Era of Cloud Native Computing Power" was held at the Yunqi Scientific and Technological Innovation Show in Hall D of the Hangzhou Yunqi Conference. He Yaming, general manager of Alibaba Cloud Intelligent Video Cloud, brought a theme sharing entitled "A New Paradigm of Video Cloud Native Business Based on Relying on Heaven", which introduced in detail the new trend of video cloud native business based on Relying on Cloud Server. This article is compiled according to the speech for everyone to read.

01 The first market share, Alibaba Cloud video cloud edge capability highlights

As a form of content expression, video has now covered many industries such as live broadcast and short video. The new era of social video, that is, "social video", has arrived. In the content output dimension, there are not only pre-filming, but also video processing, distribution, dissemination and other actions, and AI functions are added.

Alibaba Cloud Video Cloud's cloud-side integrated digital intelligence audio and video capabilities include the advantages and features of cloud-side integration, elastic scalability, smooth interaction, technology inclusion, digital intelligence upgrading, etc. Therefore, we believe that video is a technical field created by interdisciplinary disciplines. Whether it is traditional signal processing, encoding and decoding, transmission, network, AI, etc., a large number of computing needs must be the first choice.

In the IDC report, Alibaba Cloud Video Cloud won the first place in the overall market share of China's video cloud for four consecutive years. It has also been mentioned earlier that the video is directly uploaded to live broadcast and distributed after the shooting is completed, but the actual technical processing needs to go through a very long link, and the cost, threshold and professionalism are very high.

The reason why video cloud is the best case of cloud native is that the entire video cloud is "long" on Alibaba Cloud native. From top to bottom, we can see that both PaaS and SaaS tools are essentially service-oriented. At the elastic level, there are not only time and space elasticity, but also many CPU, GPU and other Intel and ARM heterogeneous resources.

02 Moving towards multiple trends, business scenarios+customer needs drive video HD technology forward

The development of video shows many trends, such as high-definition, low latency, intelligence and specialization.

Take the high-definition dimension as an example. Because of the user's sensory requirements for the ultimate experience of high-definition video, the standard definition resolution of similar video platforms has increased from the minimum SD to 1080P, which is nearly seven times the increase of information volume. To the current 4K and 8K, the improvement of resolution has brought about the expansion of a large amount of data volume. In addition to the normal needs of users, it has brought varying degrees of pressure on computing power, storage and network transmission.

At the same time, in some high motion scenes, the higher the video frame rate, the clearer the visual experience will be, and the same is true for wide color gamut. In the scene of immersive experience, the higher the definition and resolution of the video, the stronger the sense of immersion. These real challenges are also driving the video HD technology forward.

The second trend is low latency. When watching the live broadcast of the same football match, the next door may have been cheering, and the players you watch have not scored yet, which is the delay. Even the standard broadband TV has a delay of more than 5s. However, with the rise of immersive experience such as VR and AR, which have high requirements for low delay, the requirements for the underlying computing power and chips are also increasing.

Whether it is low latency or high bit rate, the core point is video coding. Each time the video standard is upgraded, the technical complexity will increase ten times; The quality code rate is 50% lower for each generation of video coding upgrade, but the amount of video upgrade data is more than 2 times, which is the second contradiction; At the same time, Moore's Law is slowing down, and the growth of computing power cannot keep up with the HD/8KVR live video in real time, so the third contradiction is between complexity and video computing power.

We believe that solving the three contradictions is actually a combination of three directions, namely, software and hardware cooperation to improve speed, AI processing to improve image quality, and cloud joint guarantee experience.

Software and hardware collaboration should make full use of the CPU's ability, and combine with hardware in computing, coding optimization, and computing power improvement. The computing power improvement is the main direction. Next is intelligent processing. We use a lot of AI technology to intelligently ignore the places that the human eye does not pay attention to in a large number of images. The last point is cloud collaboration. Computing power is not only available on the cloud. We can do post-processing for cloud collaboration. When computing power is not enough, we can send small resolution video. We can use the computing power on the end to make super division and make large video. All of these are our intelligent coding solutions. Of course, today we focus on "software and hardware collaboration".

03 Video cloud native business embraces Reliance ECS, which is reliable, stable and flexible

Yitian Cloud Server can play the advantages of reliability, stability and flexibility under video cloud native services, including live broadcast, on-demand, media processing, etc. Many services can be mixed, and time complexity and algorithm complexity can also be combined.

There is a constant demand for computing power. Many activities have different requirements for computing power at peak hours. Relying on the chip can give video cloud a better choice. Some new products have overclocking, frequency reduction and other problems, which make the business more complex and jittery. The selection of ARM chip combined with Yitian brings us more stability. We choose to deploy computing power in different pools in a distributed way, which also increases the escape ability of the business when the real "disaster" comes.

At the algorithm level, the combination of self-developed S263 and S265 with ARM chip has brought many efficiency improvements. In the process of making the engine, we found that the adaptation process of ARM on various video, subpackage, file, streaming media and other protocols is very flexible. At the same time, the architecture of ARM chip also brings more choices to the transcoding business of video cloud.

In terms of scheduling, after using the ARM chip, we measured the resource water level on the single machine, and the CPU utilization rate can be increased to 95%, helping the video cloud improve the CPU utilization rate by 10%.

At the level of video coding technology, ARM provides powerful tools to help the instruction set realize easy migration. Whether it is highly concurrent instructions or process optimization, it has natural advantages in architecture design, and it is also a good combination with video cloud coding.

As can be seen from the above figure, after Heping Togo cooperates with a large number of fast algorithm optimization, S264 and S265 can increase by 31.9% and 33% respectively compared with C7 after being migrated to Arm.

The improvement of performance has brought cost savings and delay reduction. Video cloud is a typical large-scale video processing scene business, and cost and experience are the core competitiveness of enterprises.

Yitian's super computing power also makes it possible to live 8K video in real time.

In terms of AI, we are constantly improving performance in terms of reasoning and adaptation compared with the x86 architecture in the past, and also positively helping video cloud reduce business costs. At the same time, the ARM model hybrid has also been launched in the video cloud on-demand business.

04 Rely on the ECS+Arm chip to give full play to the advantages of computing power and build core competitiveness

The demand of customers and business scenarios brings more challenges to video technology. Whether it is clearer, smoother, lower latency or higher definition experience, we need to provide strong support for video in terms of computing power. In the future, we will deeply cooperate with Yitian Cloud Server and ARM chip to play the advantage of computing power.