From April 16 to 17, 2021, LiveVideo PackCon, the largest multimedia technology summit in China, was held in Shanghai. This conference on “new technology, new opportunities” as the theme, focusing on the latest audio, video, images, and other technical exploration and application practice, topics include the education, entertainment, health care, financial, social, game, smart devices, and other areas of the industry, brings together audio and video field of top technical experts at home and abroad, attracted nearly thousands of audio and video developers. As a leading technology audio and video manufacturer in the industry, Bileyun was invited to share technology, and shared the best Practice of Audio and video System Congestion Control for many guests in the special conference of “Network Transmission and RTC”. The whole dry goods shocked the audience.

With the development of multimedia technology, application scenarios and coverage become more and more extensive. Cloud gaming, UHD video, AR/VR…… For network transmission, it represents severe challenges such as higher bandwidth and lower latency. In the topic sharing of network transmission, Volvet, chief scientist of Beiluyun, talked about network congestion. He pointed out that: Similar to road traffic congestion, network congestion is the deterioration of network service quality caused by the amount of data carried by network nodes and links exceeding its capacity. Network congestion is often accompanied by packet discarding. In real-time audio and video systems, packet loss is the most direct cause of audio and video experience decline. There are no traffic police directing traffic on the Internet, so we need a way to guarantee the quality of network service under the changeable network conditions, which is congestion control.

In the real-time audio and video system, the quality requirements can be summarized as two high and one low, namely: high fluency, high clarity, low delay. Standard G.114 of itu-t provides some guidelines for one-way delay in transmission, which is generally considered to be less than 400ms for RTC systems. The clarity and fluency of audio and video are closely related to the scene, and different scenes have different requirements for clarity.

Volvet believes that ensuring good communication links is the key to a good audio and video experience. The first step is to find the right highway entrance and choose the right access to the data center. There are two common methods: the local proximity principle is adopted in global scheduling, or the client selects the optimal link to establish a link. The second step is dynamic routing planning. Sometimes the shortest distance between two points may not be a straight line, so the best route can give consideration to audio and video communication quality and server resource loss. PANO Backbone (global Real-time transmission Acceleration network) constructed by Paileyun is a distributed system with multi-DC multi-level scheduling, which achieves the best effect on link optimization.

According to Volvet, the goal of congestion control is to minimize packet loss and jitter caused by congestion, and network evaluation model plays a crucial role in this regard. However, in practical scenarios, weak networks cannot be completely avoided. In the means to resist weak networks, packet loss retransmission and forward error correction coding are commonly used to resist packet loss. The JitterBuffer is used to resist network jitter. The principle of alchemy in The Japanese book The Alchemist of Steel is equal exchange, and all means to resist the weak network need to pay the price, which can also be regarded as equal exchange. From this perspective, Occam’s Razor principle and NFL (No Free Lunch) principle are also applicable to guide the design of congestion control algorithm.

Our core technical team has been focusing on audio and video development for nearly 20 years, and has reached the international top level in audio and video coding and decoding, network transmission, weak network confrontation and QoE, echo cancellation, real-time communication networking and routing, high-concurrency streaming media distribution and other aspects. The current product matrix includes: Voice calls, video calls, interactive whiteboard, interactive live, etc., offer a variety of whole platform native SDK and cross platform SDK, businesses and developers can quickly implement interactive worldwide small-class, super small classes, double normal university class, voice chat, video, social live, even the wheat, games, audio, video, remote medical service, such as office collaboration scenarios.

In the future, based on continuous breakthroughs in technology, Paileyun will explore the pain point solutions of industry users in multimedia communication scenarios, provide developers with more product innovation and imagination space, and provide users with better audio and video experience.