What Makes Kuaishou’s CMTP the Future of Media Transmission?

Audio‑Video Era and Media Transmission Protocols

Audio‑video applications have surged in recent years, becoming a major carrier of internet traffic with diverse formats and numerous transmission protocols. At LiveVideoStackCon 2022 Beijing, Kuaishou’s transmission algorithm lead Zhou Chao presented the evolution of media transmission protocols, challenges, and the new CMTP standard.

Classification of Media Transmission Scenarios

Based on user experience, media applications can be divided into three categories: generic VoD, generic RTC, and generic Live. Generic VoD focuses on reliability and bandwidth utilization, generic RTC prioritizes ultra‑low latency, and generic Live balances latency, reliability, and bandwidth according to specific live‑streaming scenarios.

Architecture of Generic VoD

VoD processing includes video capture, transcoding, cloud upload, server‑side review and preprocessing, secondary transcoding for higher compression, and CDN distribution for download, decoding, and rendering.

Architecture of Generic Live and RTC

Live streaming shares a similar pipeline but must handle real‑time generation, requiring stable transmission; compromises on bandwidth and latency are made, with protocols such as RTMP, QUIC‑based RTMP, and emerging WebRTC solutions. RTC aims for sub‑second interaction latency, typically using WebRTC with custom enhancements.

Kuaishou’s Practice in Media Transmission Optimization

Kuaishou’s transmission stack relies on core algorithms (congestion control, multi‑bitrate adaptation, weak‑network resistance) and a suite of proprietary protocols: KTP (private protocol for live push, upload, RTC), LAS (low‑latency live multi‑bitrate adaptation, now an industry standard), KLP (live pull protocol), and AAS (VoD multi‑bitrate adaptation).

KTP is designed to support VoD, Live, and RTC simultaneously, balancing latency, reliability, and bandwidth utilization through a two‑layer architecture: a transport control layer and a business‑aware layer.

Performance tests show KTP maintains clear, smooth streaming at 60% packet loss, whereas RTMP fails at 15% loss.

In upload scenarios, KTP‑based services raised success rates from 70‑80% to over 99% and reduced upload latency despite growing file sizes.

KTP also powers all Kuaishou RTC services (PK, link‑mic, meetings, StreamLake), delivering superior experience and performance over competitors.

Kuaishou’s congestion control algorithms (IA2C, NNCC, upcoming AQDC) outperform BBR, achieving higher bandwidth utilization and lower latency.

Next‑Generation Media Transmission Protocol: CMTP

Building on previous protocol successes, CMTP aims to be a universal protocol for all scenarios, offering architecture universality, full‑scene coverage, high extensibility, and rich features.

CMTP consists of five layers:

UDP/TCP Layer : Default UDP for flexibility; fallback to TCP when UDP is blocked.

Transport Control Layer : Supports UDP/TCP, defines fields, packetization, session management, ARQ, FEC, congestion control, 0‑RTT, encryption, multiplexing.

Transport Presentation Layer : Defines media session, stream, data, and signaling representation, with protocol selection.

Application Awareness Layer : Componentized business‑aware modules for Live, VoD, RTC, and generic use, plug‑and‑play and extensible.

Common Interface Layer : Standard external interfaces and configurations for client/server, metadata, and generic settings.

CMTP has already been deployed at Kuaishou with significant benefits, and many vendors now support it, advancing its standardization and ecosystem.