High‑Availability Architecture and Scaling Practices of Meipai Short‑Video Platform

1. Development of the Short Video Market

In recent years short‑video applications have exploded in China; Meitu launched Meipai along with competitors such as Kuaishou, Miaopai, etc., enriching the market.

The rapid growth is driven by increased bandwidth, improved mobile hardware, richer expressive capability of video, and product features that lower creation barriers.

2. Growth of Meipai

Meipai was released in May 2014, topped the App Store free chart on day one, reached 100 million users within nine months, and now sees daily video playback exceeding 270 million views and 1.83 million hours.

After a year of architectural evolution, Meipai has accumulated experience in building a highly available and scalable system.

Compared with text‑based social apps, a short‑video product faces specific architectural challenges that will be discussed.

3. Architectural Challenges of Short Video

Data size : A 10‑second video is about 1 MB, far larger than text, leading to challenges in upload, storage, and playback. Large files require CDN acceleration, chunked upload, and multi‑bitrate adaptive streaming (HTTP range or HLS).

Data format : Video uses binary codecs such as H.264/H.265, requiring standard handling.

Processing demands : Watermarking, thumbnail generation, transcoding are resource‑intensive. Meipai splits processing between client (using soft/hard encoding, face‑recognition, beautification) and server (ffmpeg‑based transcoding, moderation).

4. Improvements for Billion‑User Scale

Performance, availability, burst traffic, and rapid iteration are the main challenges.

The overall architecture (illustrated below) continuously evolves, focusing on service governance, modularization, and isolation.

Two communication methods are used: HTTP (for cross‑team, cross‑language calls with Nginx load balancing) and a config‑service + RPC approach based on etcd for dynamic service discovery, load balancing, health checks, and tracing.

5. Isolation and Redundancy Strategies

Isolation is applied at multiple levels: core vs. non‑core services, intra‑cluster isolation, physical resource isolation, and external dependency isolation.

Resource redundancy includes multi‑CDN and multi‑cloud‑storage providers, with automatic failover and DNS‑based switching.

Cache redundancy follows a multi‑level design (L1 LRU, sharded caches, master‑slave) to protect databases during spikes.

6. Future Directions

With growing bandwidth pressure, a hybrid P2P + CDN solution and adoption of more efficient codecs such as H.265 are being explored.

Additional video beautification algorithms are being moved to cloud services, requiring GPU/CPU SIMD acceleration and elastic scheduling of processing clusters.