How Tencent Scaled QQ Space Video Playback to Billions: Architecture & Optimization Secrets

Wang Hui, R&D director at Tencent, presented the technical practices behind scaling QQ Space short‑video playback from 50 million daily plays to over a billion within a year.

Quick Launch

The team had to support four codebases (iOS and Android for QQ and QQ Space) and introduced a local proxy to manage video download, enabling edge‑side buffering, traffic control, pre‑loading, priority scheduling, and comprehensive monitoring.

Cost Optimization

After a 4× increase in playback, bandwidth rose 6×. They introduced multi‑level rate limiting, peak‑hour pre‑download caps, and adopted H.265 encoding, which reduced bandwidth consumption by about 31% despite higher CPU usage.

Second‑Open (Instant Start) Optimization

Targeting sub‑second start‑up, they optimized MP4 MOOV placement, implemented parallel HLS segment downloads via the proxy, and refined pre‑loading strategies, cutting average start‑up time from 1.8 s to 0.6 s.

Monitoring & Error Reporting

A detailed instrumentation framework logs each playback step, error code, buffer events, and network metrics, enabling real‑time alerts and automated root‑cause analysis; success rates improved from 97% to 99.9%.

Live Streaming & HLS

For HLS live streams, parallel segment fetching reduced start‑up from 6 s to 1.6 s, and adaptive bandwidth control further minimized stalls.

Cache & Seek Optimization

File‑hole based caching reduced seek‑induced buffering, lowering buffer probability from 4.6% to 0.48%.

Overall, the combined optimizations achieved a 99.9% download and playback success rate, sub‑second start‑up for 80% of videos, and significant bandwidth savings.