Technical Salon Summary: Tencent Cloud Video Cloud, Real‑Time Audio/Video, and Mobile Live‑Streaming Practices at 58.com

Background

On June 14, 2019, the 58 Group Technical Salon (Session 13) was held at the Beijing headquarters, jointly organized by the Technology Engineering Platform and the HR "Magic Academy". Speakers included Tencent Cloud senior architect Zhang Yuantao, senior engineers from 58's Basic Experience and Mobile Technology departments, who discussed Tencent Cloud video services, real‑time audio/video practice at 58, and mobile live‑streaming implementation.

1. Tencent Cloud Video Cloud

Tencent Cloud has built a comprehensive video ecosystem offering live streaming, video‑on‑demand (VOD), real‑time audio/video, and ultra‑high‑definition services. These solutions are widely used in education, healthcare, finance, entertainment, and other industries.

Live Streaming – Supports rich features such as text interaction, bullet comments, star likes, and co‑hosting. Multiple protocols (RTMP, HLS, FLV) are available, backed by over 1,100 acceleration nodes nationwide with 100 TB+ bandwidth and 1.6 TB/s peak bandwidth, serving platforms like Douyu and Longzhu.

VOD Integration – Provides end‑to‑end storage, transcoding, and accelerated playback. Tasks are launched synchronously and results are notified asynchronously, with customizable task flows and anti‑leech protection.

Real‑Time Audio/Video (Tencent‑RTC) – Based on years of QQ experience, it offers cross‑platform high‑quality video calls for mini‑programs, H5 pages, apps, and PC clients, suitable for online education, insurance assessment, tele‑medicine, and micro‑policing.

Ultra‑HD – Uses deep‑learning models to classify video scenes (game, indoor, outdoor, sports, short video, animation) and applies scene‑aware encoding parameters, ROI‑based denoising, and dynamic bitrate adjustment for optimal quality at lower bandwidth.

2. Real‑Time Audio/Video Practice at 58

Real‑time communication is built on the WebRTC stack, supporting APP, PC, and mini‑program clients. Signalling is carried over IM (WeChat‑like) rooms, with NAT traversal via relay services. Media is routed through signalling‑conversion and media‑relay services.

IP‑Phone Architecture – To reach users without the app, WebRTC media is converted to SIP signalling and RTP media, enabling traditional phone calls while preserving the same backend.

Business Integration – Audio/video calls are deployed in 58 APP, Ganji, Zhaocaimao, Anjuke, Zhuanzhuan, etc. Specific use cases include:

Second‑hand car sales: start with video call; if the dealer is offline after 4 seconds, automatically switch to IP‑phone to the dealer’s mobile.

Recruitment: switch from P2P to PSP mode, routing streams through the server for recording.

3. Mobile Live‑Streaming SDK Practice at 58

The 58 Live‑Streaming SDK provides a one‑stop solution for mobile live‑streaming across various business lines (real‑estate, community, Anjuke, mobile agents, etc.). It includes modules for image/audio capture, processing, encoding, FLV packaging, RTMP pushing, and chat.

Capture – Image capture uses camera preview resolution (must be ≥ push resolution) with configurable resolution, frame rate (15‑24 FPS), and YUV formats (I420, NV12, etc.). Audio capture supports 44.1 kHz, mono/stereo, 16‑bit depth.

Processing – Image processing (cropping, watermark, filters, beauty, Gaussian blur, etc.) is implemented with OpenGL ES; audio processing includes noise reduction, echo cancellation, and AGC.

Encoding – Video uses H.264 (software via open264 or hardware via MediaCodec/VideoToolbox); audio uses AAC. Both soft and hard encoding are supported, with trade‑offs in CPU usage and device compatibility.

Packaging & Transmission – Uses FLV (RTMP) and MPEG‑2‑TS (HLS) containers. RTMP offers low latency but is TCP‑based; HTTP‑FLV and HLS provide fallback with higher latency (1‑3 s for FLV, ~10 s for HLS).

Common Issues & Solutions

Pixelation/Mosaic: caused by low light, low bitrate, low frame rate, small GOP, improper profile settings; adjust resolution/bitrate accordingly.

Screen/Green artifacts: due to missing keyframes or encoder width/height not being multiples of 16; ensure proper encoder configuration.

Latency, stutter, AV sync: stem from capture/processing delay, encoding delay, network conditions, or timestamp errors; SDK dynamically adapts bitrate and resolution.

Summary

The salon highlighted the significance of video and real‑time communication technologies for 58’s business, showcasing optimization strategies, deep‑learning‑driven content‑aware encoding, and seamless integration of audio/video and IP‑phone calls to enhance user connectivity.

Outlook

Future work includes adding H.265/VP9 support, deeper AI integration for content processing, and migrating live‑stream transport to QUIC to further reduce latency and improve interactive experiences.