HUAWEI CLOUD, based on cloud-native media network, releases new heavyweight products

Abstract: Huawei Cloud TechWave Cloud Native Media Service Special Day, Lu Zhenyu, Cloud Video Director of Huawei Cloud Media Service Product Department, delivered a speech "Cloud Native Media Network, Upgrading Tradition, and Empowering the Future", sharing why we need to restructure the media network. What are the current problems that a media network built based on Huawei Cloud’s cloud native technology can solve, and why cloud native is a necessary option for future media networks.

This article is shared from the HUAWEI CLOUD community " HUAWEI CLOUD based on cloud-native media network, another blockbuster new product ", the original author: Audio and Video Manager.

On June 25th, Huawei Cloud TechWave Cloud Native Media Service Special Day was successfully held online. Huawei Cloud audio and video services are officially upgraded to media services. The upgraded media service supports the three major scenarios of media production, media distribution and media application. On the special day, Lu Zhenyu, Director of Cloud Video, Huawei Cloud Media Service Product Department, delivered a speech "Cloud Native Media Network, Upgrading Tradition, and Empowering the Future", sharing why media networks need to be reconstructed, based on the cloud native technology of Huawei Cloud. What are the current problems that media networks can solve, and why cloud native is a necessary option for future media networks. The following content is organized according to the content of the speech, and there are deletions.

Challenges and opportunities coexist: Traditional networks are no longer adapted to the surge in traffic and new business forms

In the past few years, we have seen that with the progress of mobile Internet and 5G, global traffic, especially video traffic, has increased by 12 times within 5 years. We have seen a lot of 1080p, 4k, 8k, and VR videos, making the network traffic and bandwidth rapidly burst.

Traditional CDN and live broadcast can well solve the distribution of on-demand video and traditional live video, but the usage scenarios are limited, because there is a delay of 3 to 5 seconds, and they cannot support two-way interactive, real-time interactive video distribution scenarios. For example, ultra-low-latency live broadcast, real-time audio and video, and some video surveillance scenarios, the form of these services puts forward higher requirements for the ability of media distribution.

In the live streaming scene, it is often the anchor who starts to introduce the product, and the user terminal shows that the link is already up, but after 5 seconds, the user hears the anchor saying that the countdown has started, and the link is on 5, 4, 3, 2, and 1, if At this time, the user clicks on the link again and will find that the product may have been grabbed, because there is a 5 second delay between the live broadcast and the text. In such a scenario, ultra-low latency live broadcast within 1 second is required, which is not available in the current media distribution capabilities. We call "too late" .

Another scene is "Can't afford to use" . Everyone hopes to connect the content in the video camera to the cloud at low cost and quickly, because there are a large number of AI capabilities on the cloud, and there are many business innovations that can make video create greater value. Calculating with the current model: Take an ordinary small park as an example, the bandwidth of each channel of 100 cameras is 1 megabyte, which is 100 megabytes. The cost of such a cloud broadband access will be more than 120,000 yuan per year. This is everyone Far unbearable. Limited by unaffordable costs, many parks can only store these video content offline. The problem is that these video content assets cannot be realized well and cannot create more value.

Say goodbye to traditional networks: A cloud-native media network with integrated media nodes, hierarchical design, and AI scheduling features emerges as needed

In order to solve the problems of "too late" and "cannot afford", Huawei Cloud proposed a cloud-native media network. We believe that cloud-native media networks should have three characteristics: converged media nodes; three, four, and seven layers Layer network transmission optimization; AI intelligent scheduling.

Converged Media Node In the past, we had CDN services, CDN nodes, live broadcast services, and live broadcast nodes. Each business of RTC and video surveillance has its own nodes. These nodes cannot be shared and cannot be flexible. The deployment of services will be subject to many restrictions.

The first thing HUAWEI CLOUD did was to transform the 2500 nodes around the world in a unified cloud-native way, and upgrade them into converged media edge nodes. Its feature is unified computing and storage resources at the bottom layer, and native nodes are used on it. The IEF of China has undergone unified transformation and management, so that CDN, live broadcast, RTC, XR, transcoding, etc. are all services that are dynamically deployed based on cloud-native capabilities.

Each service can be scheduled during the day and night according to its own needs, using uplink and downlink bandwidth and computing and storage resources. In this way, edge assets can be activated, which can greatly reduce service transmission delays and more important service costs.

layered design We use layered thinking to optimize network transmission: similar to TCP/IP, we need to consider audio and video content. When transmitting on the network, it encounters packet loss, delay and different network access conditions The problem of different parameters is solved by a layered idea, which is divided into 3 layers, 4 layers and 7 layers for optimization. At the third layer, which is the IP layer, we proposed Tianlu, which can perceive the network in real time, optimize forwarding and routing, improve the delay of packet forwarding, and increase the arrival rate of packets. After actual measurement, we can reduce the forwarding delay of IP network and IP packets by 30%, and increase the arrival rate of packets by 0.5 percentage points.

Also on the 4th layer, which is the transport layer, Huawei's self-developed hQUIC protocol is proposed, so that upper-layer application development does not need to perceive what kind of network medium the lower layer is running on, whether it is WiFi, 5G, or other networks. It can enable the content of audio and video services, real-time message transmission services, and future XR cloud game services to be accelerated separately under different network conditions, and improve the efficiency of network transmission.

On the 7th layer, the problem of audio and video service transmission is solved in various business scenarios. For example, do you need to perform some encoding and decoding first, and then transmit to improve efficiency? Do you need to adapt some codec parameters? Is there a better experience in the middle of transmission... These are all done on the 7th layer. Through layering, we optimize at layers 3, 4, and 7, so that the media network can provide a very consistent experience for different audio and video content on different networks.

AI Scheduling We have proposed a Mesh-based scheduling engine, which is the result of our cooperation with domestic Top1 schools. It has two key features. The first feature is the unified scheduling of multiple services. Multiple services include on-demand, live broadcast, RTC, monitoring, etc. We can complement and optimize experience and cost to achieve end-to-end unified scheduling of multiple services Ability, different services are uniformly scheduled by this scheduling engine.

The second feature is that different service SLAs adopt different scheduling strategies to support the business strategies of different customers. For example, for live broadcasts, do we pay more attention to the return-to-source rate? , Bandwidth trend, stall rate and other cost and experience indicators, we use these indicators as an input to the scheduling engine. Among 2500 nodes, we always select the best node for dynamic deployment, elastic scaling and the most Excellent service.

For scenarios such as RTC with higher real-time, stronger interaction and more demanding experience indicators, we will input the user’s first frame duration, number of freezes, room entry success rate, end-to-end delay, and other parameter groups into the In the scheduling engine, RTC always provides a better user experience, lower latency, and helps customers' business. In order to achieve such scheduling capabilities, we implemented Channel-level scheduling in the implementation of the scheduling system.

We perform overall scheduling and optimization of the video access phase, return-to-source phase, and the entire network according to a Mesh network, and introduce artificial intelligence to self-learn, self-train, and continuously train the scheduling algorithm. Confrontation and generate better scheduling strategies and algorithms. Through the unified upgraded media edge network, our AI scheduling, and the optimization of the layer 3, 4, and 7 network transmission layers, our cloud-native media network can solve the problem of "too late" and "cannot afford".

Let me give you two examples. The first example is on such a media network. We can upgrade the live broadcast service to ultra-low latency live broadcast. The current live broadcast has a delay of 3 to 5 seconds. We will use RTC for live broadcast. The method is upgraded to ultra-low-latency live broadcast, which reduces the delay to less than 800 milliseconds, but other business indicators experience, the first screen freeze is still the same as the traditional live broadcast, such a service will help the live broadcast, including e-commerce, education industry Make a big experience upgrade.

At the same time, if the customer is willing to accept the transfer of its architecture station from live broadcast to RTC, we can also directly provide RTC services on such a cloud-native media network, changing the customer’s transmission from TCP to UDP, which will bring a further realization of the experience. Upgrade and delay reduction.

The second example is a cloud-native video surveillance service (VIS service), which is also a cloud-native media network that fully multiplexes the upstream traffic in the network, making the access price of the camera more affordable. From the example I just gave, reducing from 300 yuan/month to double digits per month can greatly reduce the operation and maintenance cost of each camera, and encourage users to put the content generated by the camera on the cloud for more enjoyment. With low-cost storage, and more innovation and AI capabilities on the cloud, we can make users' monthly costs less than 10% of the past.

blockbuster new product: open, shared, efficient, and secure video access service, based on the edge collaboration architecture, supporting multiple business scenarios

Today I am honored to release Huawei’s VIS service product, Video Ingestion Service, and video access service. The video access service is an open, shared, efficient, and secure video access service based on the edge collaboration architecture on HUAWEI CLOUD. It supports multiple business scenarios and video cameras, similar to the access of the national standard 28181 video protocol, RTMP video, and FLV video. We will continue to increase support for SRT and more smart cameras and video access protocols in the future.

At the same time, we will continue to expand the boundaries of services and upgrade the value of services. The content of the video returns to the cloud. Only when it collides with more industry ecology can it produce more innovation and bring more value.

HUAWEI CLOUD not only has self-developed algorithms, but also an open algorithm store. There are many third-party algorithms. These algorithms are continuously evolving and superimposed infinitely. We firmly believe that only by connecting more videos to our cloud at low cost can we create more value. Then we can watch a small video to see how Huawei Cloud Video Access Service can easily upload the content of the video camera to the cloud.

I am very happy to share with you the native media network of HUAWEI CLOUD today. We introduced the challenges we encountered and introduced how HUAWEI through the convergence of media nodes, layer 3, 4, and 7 layer optimization, and AI scheduling to form one Zhang Xin’s cloud-native media network solves these problems, and we have introduced products such as ultra-low latency live broadcast and VIS video access. thank you all.

Click to follow and learn about Huawei Cloud's fresh technology for the first time~