Meitu Short‑Video Cost Halving and Millisecond Startup Optimization Practices

In recent years short‑video has become a major internet trend. Meitu, one of the earliest short‑video apps (since 2014), has continuously invested in short‑video technology. This article describes Meitu’s practice of halving short‑video costs and achieving millisecond‑level startup by optimizing cost, success rate, and playback experience.

Technical Background

With the development of 4G and broadband, users have shifted from text to rich streaming content, raising expectations for video quality and latency. Meitu identified three main pain points in the early stage:

Poor viewing experience – a 1‑3 s spinner before playback and sometimes the need to download the whole video before playing.

Slow startup – videos are only a few seconds to minutes long, and a 1‑3 s start delay reduces watch‑through.

High download cost – CDN bandwidth dominates video‑service expenses.

Data‑Driven Optimization

Meitu built an objective data system to guide all optimizations. The system emphasizes accuracy, rapid rollout, and reproducibility, enabling:

Evaluation of optimization impact.

Real‑time fault alerts.

Quality‑based decision making.

User‑issue troubleshooting.

Fusion Scheduling System

To address core pain points, Meitu designed a unified on‑demand scheduling platform consisting of:

Scheduling service – multi‑CDN and multi‑bitrate dispatch with extensible policies.

Quality system – client‑reported metrics processed in real time to drive automatic scheduling decisions.

Cloud processing – analysis of hot files to generate multi‑bitrate (e.g., H265) assets for flexible delivery.

Cost Optimization

The primary goal is to reduce CDN bandwidth while preserving visual quality. Meitu assumes that many users abandon playback early, so caching the initial segment locally can avoid repeated downloads. Strategies include:

Rate‑limiting based on bitrate information to keep download speed aligned with playback bitrate.

Controlling player buffer length so that only the necessary portion of a video is fetched before a potential quit.

These measures cut overall bandwidth cost by more than 50 % without degrading picture quality.

Success‑Rate Improvement

Multi‑CDN deployment with automatic failover reduces regional outages.

Reducing unnecessary network requests lowers error probability.

Integrating FastDNS improves DNS resolution success and cuts latency by 60‑100 ms, decreasing request failure rate by ~40 %.

Combined, the error rate dropped from ~1 % to 0.1 % and latency was reduced accordingly.

Playback Experience Optimization

Key techniques:

Fast‑start (moov relocation) : Moving the MP4 ‘moov’ box to the file head enables immediate metadata access, eliminating the need to download the entire file before playback.

Player tuning : Adjusting ijkplayer’s FAST_BUFFERING_CHECK_PER_MILLISECONDS from 500 ms to 50 ms saves ~300 ms of startup time.

Pre‑batch scheduling : Leveraging the time gap between feed refresh and video start to cache scheduling results, offsetting the extra 300 ms introduced by the fusion scheduler.

Pre‑loading : Loading a small (~600 bytes) moov segment in advance for videos likely to be played, achieving 45 % of videos starting within 250 ms and 85 % within 750 ms.

Summary

CDN bandwidth reduced by >50 % with unchanged visual quality.

One extra range request eliminated, saving 100‑200 ms.

FastDNS saves 60‑100 ms.

Fusion scheduling adds ~300 ms but cuts error rate from 1 % to 0.1 %.

Player parameter tweaks save ~300 ms.

Batch pre‑scheduling saves 200‑300 ms.

Pre‑loading brings 45 % of videos to start within 250 ms.

Meitu continues to explore deeper optimizations such as quality‑preserving compression, narrow‑band HD, and emerging AI‑driven techniques.