Technical Overview of Tencent Cloud X‑P2P Low‑Latency Video Streaming and P2P Technologies

On June 29, a technical salon on audio‑video and convergent communication concluded with a presentation by Tencent Cloud senior engineer Zhang Peng, who leads the X‑P2P live‑stream acceleration project.

Zhang introduced his background (Huawei University graduate, experience in game development, operations systems, and video P2P research since 2014) and outlined the evolution of P2P from early RFC concepts, through the golden era of academic projects (CoolStreaming, AnySee, GridMedia), to its decline and recent resurgence with live‑streaming.

The talk explained why modern P2P is needed for video: bandwidth costs remain high while user expectations for low latency and smooth playback increase. Tencent Cloud X‑P2P achieves share rates of 60‑80% and latency comparable to or better than CDN.

Key technical points covered:

Data consistency for VOD (offset‑based slicing) versus live streams (DTS‑based slicing).

Traditional CDN‑style segmenting (HLS/DASH) versus X‑P2P’s ability to segment raw FLV/FMP4 streams without altering mux or codec information.

Adaptive bitrate on FLV/FMP4 by aligning PTS and inserting new I‑frames, allowing seamless bitrate switches without breaking the player.

Network traversal was detailed, describing NAT types (public, firewall‑blocked, symmetric firewall, full‑cone, restricted cone, port‑restricted cone, symmetric) and the role of STUN in establishing peer‑to‑peer connections. Tencent’s experience with symmetric NAT traversal, built on QQ/WeChat technologies, was highlighted.

Three P2P topology models were compared:

Mesh model – high share rate but higher latency and signaling overhead.

Tree model – efficient distribution but vulnerable to node failures and leaf‑node under‑contribution.

Balanced (average‑share) model – peers grouped to evenly distribute pieces.

The XNTP transport protocol was introduced as an improvement over TCP and TFRC. XNTP features fast start (detecting ideal bandwidth within one RTT), fair bandwidth sharing with TCP, better congestion control, reduced retransmission ambiguity, and pacing to avoid bursty traffic.

Future outlook includes the impact of 4K, 5G, IPv6, blockchain’s P2P concepts, edge computing reliance on robust P2P, and the possibility of X‑P2P approaching QUIC‑level performance, ultimately moving toward a decentralized internet service model.