Practice and Principles of User Access Experience Optimization for Vivo Short Video

This article, authored by the Vivo Internet Operations team, introduces the practice ideas and underlying principles for optimizing the user access experience of Vivo short video.

Background : In short‑video apps, a few seconds of frozen frames often cause users to swipe away. Startup latency, which consists of video file download time and video decoding time, directly impacts user retention.

The article approaches the problem from an operations perspective, analyzing each network link in the user request chain and sharing concrete optimization cases.

User Access Chain : The complete network request process from the client side is illustrated (see diagram). When CDN is involved, the process includes DNS resolution, TCP connection establishment, TLS handshake, and CDN response.

DNS Resolution : DNS translates domain names to IP addresses. The article explains the hierarchical structure of domain names, the step‑by‑step resolution workflow, and the role of root, top‑level, and authoritative DNS servers (see diagrams). It identifies unstable DNS resolution latency for Vivo short‑video domains, caused by low cache‑hit rates in certain regions.

Optimization: To increase local DNS cache hit rate, a nationwide DNS‑warming task was added, resulting in roughly a 30 ms reduction in DNS lookup time (see before/after chart).

HTTP Performance : The article compares HTTP/1, HTTP/2, and HTTP/3. It highlights the drawbacks of HTTP/1.0 (per‑request TCP handshake), the improvements of HTTP/1.1 (persistent connections, pipelining), and the remaining bottlenecks such as uncompressed headers and head‑of‑line blocking.

HTTP/2 introduces header compression (HPACK), binary framing, streams, multiplexing, and server push, which greatly reduce latency and improve connection utilization. However, because HTTP/2 still runs over TCP, it inherits TCP‑level head‑of‑line blocking, handshake latency, and connection migration issues.

HTTP/3 replaces TCP with QUIC over UDP, eliminating head‑of‑line blocking, reducing handshake to a single RTT (or 0‑RTT on repeat connections), and supporting seamless connection migration via connection IDs. QUIC also uses QPACK to avoid HPACK’s head‑of‑line blocking.

Vivo short‑video optimizations include:

Using HTTP/1.1: Merged first‑frame image domain and comment avatar domain, increasing TCP reuse by 4 % and reducing average image load time by ~40 ms.

Adopting HTTP/2 on selected domains, decreasing video stutter rate by 0.5 %.

Deploying QUIC for weak‑network scenarios and further tuning QUIC for short‑video workloads.

CDN Acceleration : CDN distributes static resources to edge nodes close to users, reducing long‑distance latency. The article explains pull vs. push caching, the role of Global Server Load Balancer (GSLB) in selecting the nearest CDN node, and the DNS redirection flow (see diagrams).

Vivo’s CDN optimizations involved adjusting edge node placement for regions with poor coverage, cutting average request latency to ~300 ms and first‑byte latency to ~100 ms.

Summary and Outlook : Continuous user‑experience optimization is essential as the service grows. Additional explored improvements include connection reuse for video scrolling, pre‑caching files to lower startup cost, and content size reduction.