Achieving Near‑Zero First‑Frame Delay in Video On‑Demand

VOD End‑to‑End Audio‑Video Solution

The Volcano Engine VOD end‑to‑end solution covers the full video lifecycle from upload to playback, consisting of four main modules: Upload SDK , Video processing and management , CDN distribution , and the VOD SDK on the playback side. Each module has been continuously optimized to meet growing user and business demands.

Playback Quality Metrics

User experience is measured from three dimensions: playback source quality, interaction experience, and viewing experience. These are further divided into three layers:

QoS (Quality of Service) : technical indicators such as playback failure rate, start‑up time, stall metrics, and picture quality.

QoE (Quality of Experience) : combines QoS with business‑side data like play count, duration, completion rate, and contribution metrics.

Business Data : final business indicators such as DAU, retention, ad revenue, and cost.

Understanding First‑Frame Time

First‑frame time is defined as the interval from the user’s play action (click, swipe, etc.) to the moment the first video frame is rendered. It includes both business‑side latency (page creation, UI rendering) and player‑side latency (prepare, data download, decode, render).

The playback lifecycle is split into three states: pre‑play, playing, and completed. Each state contributes to the overall latency and is tracked via session or trace IDs for detailed analysis.

First‑Frame Decomposition

The total first‑frame latency is broken down into four components:

Business time : page creation, interaction, and rendering on the app side.

Player kernel initialization : module initialization within the playback engine.

Network time : DNS resolution, TCP connection, and receipt of the first video packet.

Decode & render time : decoding and rendering of the video frame.

“Zero‑Delay” First‑Frame Optimization Practices

“Zero‑delay” does not mean literal 0 ms; it refers to a perception‑free start where the user sees a smooth playback without a noticeable pause. In production, about 50 % of first‑frames are already under 100 ms, and human perception becomes negligible below 200 ms, making 100 ms a practical target for zero‑delay.

Business Time Optimization – Pre‑Rendering

Pre‑loading alone does not tightly couple the player with network I/O. The solution is to pre‑render the next video by creating an additional player instance that prepares the first frame while the current video is still playing. This reduces the perceived first‑frame delay by replacing the static cover with an actual video frame.

Network Time Optimization – Node Selection

Before the first frame arrives, data preparation involves DNS lookup and IP selection. Node selection chooses the optimal IP from a pool to minimize latency, avoid overload, and improve reliability. Strategies include blacklisting problematic IPs, attributing request failures to either client or node, and ranking nodes based on QoS/QoE metrics.

Network Time Optimization – Decoding Time

Decoding latency is a key advantage of the Volcano Engine player. The decoding pipeline includes format parsing, demuxing, audio/video decoding, and rendering. Optimizations applied:

Parallel decoder initialization : header parsing and codec initialization run in parallel with data download, saving 80‑120 ms.

Decoder reuse : a codec pool reuses the previous video’s decoder for the next video, cutting another ~40 ms.

Device‑capability big‑data selection : large‑scale telemetry determines whether a device can use hardware decoding; fallback to software decoding is minimized, reducing fallback occurrences to 0.3 % of sessions.

First‑Frame Optimization Goal – Turning‑Point Analysis

Data analysis shows user tolerance thresholds: most users abandon within 50 ms, tolerance is relatively flat between 70 ms and 200 ms, and experience degrades sharply after 200 ms. Therefore, keeping first‑frame latency under 200 ms is essential for lossless experience, and sub‑200 ms allows metric trade‑offs.

Summary and Outlook

The VOD middle platform aims to deliver an extreme playback experience by balancing technical optimizations with user perception. The three‑high vision includes:

High universality : easy integration for diverse business scenarios.

High extensibility : customizable capabilities for different business needs.

High quality assurance : full‑link monitoring, fault diagnosis, and optimal VOD strategies for billions of users.