Tencent Cloud X-P2P Live Streaming Acceleration Solution

Tencent Cloud has been developing the X-P2P solution since 2014. After more than two years of research, multiple product lines have matured, covering various platforms and latency scenarios for both live and on‑demand P2P streaming. The technology is already deployed on major live platforms such as Douyu and Panda, and has withstood high‑traffic events like national parades and sports broadcasts. On November 18, senior engineer Zhang Peng will present the talk “Tencent Cloud X‑P2P Live Streaming Acceleration Solution”.

The X‑P2P product is an industry‑leading, mature P2P system. It supports different platforms and latency requirements, and has been proven in large‑scale live streaming scenarios.

P2P Technology Development History

The P2P concept first appeared in 1969 (RFC 1). In 2000, Gnutella emerged, followed by the release of the BitTorrent protocol. Until the end of 2004, when PPLive was launched, developers began to realize that P2P could also serve streaming media. From 2005 onward, P2P became widespread in live streaming. Key milestones include:

2005 : Hong Kong University of Science and Technology introduced Goo1Streaming, a mesh‑based unstructured network that periodically exchanges buffer maps and neighbor lists, achieving load‑balanced peer sharing, though it suffers from poor connections between distant nodes.

2006 : Huazhong University of Science and Technology developed AnySee, a tree‑structured live streaming solution that builds IP multicast based on IP proximity, improving neighbor quality but vulnerable to parent‑node failures.

2008 : Tsinghua University created GridMedia using a push‑data model, delivering low latency and performing well in large‑scale events such as the Spring Festival Gala and the Olympics, though performance degrades with few users.

Tencent Cloud X‑P2P Architecture and Optimizations

The architecture is segment‑based. The broadcaster pushes the media source to the server, where it is sliced into 1‑second segments and integrated into the CTN (Content Transfer Network) for origin pull. The client first contacts a COM service to obtain channel information, including penetration servers, log servers, and the latest segment list. After receiving the seed server address, the client initiates P2P connections and data exchange begins.

Three major optimization solutions were presented:

Internal Transmission Control : Nodes compete and evolve automatically, avoiding TCP resource contention and burst packet sending, thus preventing network congestion and ensuring stable playback.

Precise Playback Control : Fast‑write mechanisms enable “second‑play” (instant video start) and fast‑forward at startup to eliminate GOP delay, meeting low‑latency requirements.

Large‑Room High Concurrency : The system proactively splits large rooms, assigning a dedicated server to each sub‑room to distribute users and alleviate load.

Challenges ahead include handling higher video resolutions (4K, 10 Mbps) that outpace bandwidth growth, and cross‑province, cross‑operator traffic that may cause ISP dissatisfaction. The future direction points to P4P as a potential solution.