[Paper Review] Characterizing Internet Video for Large-scale Active Measurements

Saba Ahsan, Varun Singh+1 more|arXiv (Cornell University)|Aug 7, 2014

Image and Video Quality Assessment15 references3 citations

TL;DR

This paper presents a large-scale analysis of over 130,000 popular YouTube videos from 58 countries (2013–2014) to inform active measurement design for Internet video. It finds that video length and file size follow a lognormal distribution, and that the first 3 minutes of a video adequately represent its long-term bit rate variation and burstiness, enabling efficient, representative active measurements without disrupting user traffic.

ABSTRACT

The availability of high definition video content on the web has brought about a significant change in the characteristics of Internet video, but not many studies on characterizing video have been done after this change. Video characteristics such as video length, format, target bit rate, and resolution provide valuable input to design Adaptive Bit Rate (ABR) algorithms, sizing playout buffers in Dynamic Adaptive HTTP streaming (DASH) players, model the variability in video frame sizes, etc. This paper presents datasets collected in 2013 and 2014 that contains over 130,000 videos from YouTube's most viewed (or most popular) video charts in 58 countries. We describe the basic characteristics of the videos on YouTube for each category, format, video length, file size, and data rate variation, observing that video length and file size fit a log normal distribution. We show that three minutes of a video suffice to represent its instant data rate fluctuation and that we can infer data rate characteristics of different video resolutions from a single given one. Based on our findings, we design active measurements for measuring the performance of Internet video.

Motivation & Objective

• To understand the evolving characteristics of Internet video, especially post-HD transition, to inform active measurement design.
• To address the challenge of correlating performance across diverse video lengths and bit rates in large-scale active measurements.
• To identify a minimal, representative video duration that captures key video behavior for efficient testing.
• To enable scalable, representative active measurements by identifying correlations across video resolutions and formats.
• To support the design of adaptive streaming clients and network performance tools through empirical video behavior modeling.

Proposed method

• Collected metadata and bit rate traces from over 130,000 YouTube videos across 58 countries using location-based popularity charts.
• Analyzed video duration, file size, resolution, format (MP4, WebM), and instantaneous bit rate variation (burstiness) across categories and regions.
• Used statistical modeling to assess distributional properties, finding lognormal fits for duration and file size.
• Measured bit rate variation over time to determine the minimal representative window for burstiness characterization.
• Established correlation between bit rates across different resolutions and formats to enable upscaling/downscaling for traffic generation.
• Proposed a measurement framework where a 3-minute segment is used to infer full-video behavior, minimizing test duration and traffic impact.

Experimental results

Research questions

• RQ1 What are the statistical distributions of video duration and file size in modern Internet video?
• RQ2 How much of a video’s bit rate variation can be captured in a short initial segment?
• RQ3 Can a single video’s bit rate characteristics be used to infer those of higher or lower resolution versions?
• RQ4 To what extent do different video formats (e.g., MP4 vs. WebM) and resolutions correlate in terms of average bit rate and burstiness?
• RQ5 Can a standardized, short-duration test (e.g., 3 minutes) be used to reliably represent the behavior of longer videos in active measurements?

Key findings

• Video duration and file sizes for all resolutions (360p, 720p, 1080p) and formats (MP4, WebM) follow a lognormal distribution.
• The first 3 minutes of a video capture the majority of its long-term bit rate variation and burstiness, making it a representative window for active measurements.
• The average media bit rate for 1080p videos is typically no higher than 5–7 Mbps, with higher rates being rare.
• There is a strong correlation between media bit rates across different video resolutions, enabling traffic generation for higher resolutions via upscaling from lower ones.
• The standard deviation of instantaneous bit rates provides a reliable proxy for video burstiness, enabling video categorization for performance testing.
• File size and duration correlations across formats suggest that MP4 and WebM have comparable average bit rates, allowing either to be used for performance modeling.