How Player-Level Tweaks Can Slash Video On-Demand Costs by Up to 20%

VOD Cost Composition

In video on‑demand, CDN bandwidth accounts for about 80% of cost, while storage and transcoding together make up less than 20%; additional peripheral costs include log processing and AI processing.

Player’s Role in Cost Optimization

From 2022 to 2023, the Volcano Engine video team performed 33 cost‑optimization actions on a typical VOD service; 13 were driven directly by the player and another 12 required strong player cooperation, meaning roughly 75% of the optimizations involved the player.

In practice, player‑level improvements can save 20% or more of VOD costs.

Cost‑Optimization Relationships

Two substitution relationships exist:

Cost‑item substitution: bandwidth ↔ transcoding ↔ storage.

Cost‑experience substitution (seesaw effect): reducing waste often improves experience, and vice versa.

For example, switching from H.264 to H.265 reduces bandwidth and storage by 20‑40% but increases transcoding cost due to higher computational complexity.

Identifying Waste

Waste is defined as cached bytes that are never played; typical waste rates exceed 30%.

Unplayed exit

Backward dragging

Quality switching

Resolution overflow (e.g., 4K on a tiny screen)

Player Cost‑Optimization Methods

Cache Waste Reduction

By lowering the static cache water‑level (e.g., reducing the cache water‑mark by one‑third), the amount of wasted cached data is significantly cut.

Dynamic Water‑Level Algorithm

The algorithm adjusts cache size based on network speed, stability, and user behavior: high‑speed stable networks get smaller caches, while poor networks receive larger caches; early playback phases use lower water‑levels, later phases increase them.

Range Request Optimization

Instead of a single "0‑end" request, the video is fetched in multiple range segments; if playback stops early, the remaining segments are not requested, reducing wasted bandwidth. Two principles guide range requests: quickly fill to the target water‑level and avoid overly small range splits.

Precise Preload Strategy

Preload is refined by timing, size, and count: urgent segments (P0) are preloaded first, followed by less urgent ones (P1, P2…), with size calculated from video length, header size, and bitrate. This improves preload utilization and reduces traffic cost.

Resolution Waste Reduction

In narrow‑screen scenarios, lower‑resolution streams (e.g., 360p) are used, switching to higher quality when needed; client‑side super‑resolution (up‑scaling) further saves bandwidth while maintaining perceived quality.

Abnormal Traffic Waste

Traffic is classified into four categories based on log analysis; abnormal waste includes overly high bitrate streams caused by missing transcoding or wrong templates. A dedicated analysis module identifies and mitigates such waste.

Data‑Driven Cost Evaluation

Cost metrics are integrated into AB experiments, producing two reports: an AB experiment report (QoE impact, cost reduction) and a value‑backtrack document (long‑term savings). The core metric is "ten‑thousand‑minute playback cost" = total VOD IT cost / total playback minutes, which normalizes cost across fluctuating traffic volumes.

Cost fitting aligns client‑side logs with CDN billing data, achieving ~95% explainability.

Cost Evaluation Formula

The formula breaks down total cost into CDN, storage, transcoding, etc., and relates it to playback minutes, allowing isolation of factors such as bitrate and waste rate.

Summary and Outlook

Standard‑setting methods include benchmarking via rankings, offline experiments on an automated cost‑testing platform, and theoretical formula calculations. Consulting services combine these analyses to recommend optimal strategies for each business scenario.