Exploring the Use of Tags for Georeplicated Content Placement

Stéphane Delbruel, Davide Frey, François Taïani

Abstract

Video streaming services face significant challenges in managing global traffic, which can account for up to 60% of peak network loads. While viral content is easily managed through standard caching, the "long tail" of niche content requires more sophisticated placement strategies. This paper investigates the use of user-provided tags as proactive predictors of a video's geographic popularity. By analyzing a massive YouTube dataset, we demonstrate that a tag-based interpolation strategy can predict viewing distributions with 65.9% accuracy and improve CDN cache hit rates by up to 6.8%.

Introduction

To handle the massive volume of global video traffic, providers utilize Content Delivery Networks (CDNs) and georeplicated datacenters to store content closer to end-users. However, most current systems are reactive, caching content only after it gains popularity, or they assume a uniform global demand. The challenge lies in predicting where niche content will be popular at the moment of upload. If providers can accurately predict geographic demand early, they can proactively place replicas in the correct datacenters, thereby reducing latency and bandwidth costs.

Dataset and Metadata Analysis

The study utilizes a comprehensive YouTube dataset from March 2011, encompassing 590,897 videos and 705,415 unique tags, totaling over 173 billion views. The core of the proposed strategy relies on the metadata available at upload: tags.

Figure 5: The distribution of unique tags follows a power law, indicating a small number of very common tags and a vast "long tail" of niche tags.

As shown in Figure 5, the tag distribution follows a classic power law. While generic tags like "music" have global appeal, specific niche tags provide high-signal geographic markers. Furthermore, the availability of this metadata is consistent across the platform.

Figure 6: Most videos in the dataset contain between 1 and 20 tags, providing a sufficient metadata foundation for geographic prediction.

Figure 6 illustrates that the majority of videos are uploaded with a sufficient number of tags (typically 5 to 15) to enable effective geographic interpolation.

Methodology

The authors propose a proactive placement mechanism based on two primary concepts:

  1. Entropy Analysis: Shannon’s entropy ($H$) is used to quantify the geographic spread of a video. Low entropy indicates content concentrated in specific regions (e.g., local language content), while high entropy indicates global appeal.
  2. Tag-Based Prediction: When a video is uploaded, the system calculates a predicted geographic distribution by averaging the historical distributions of its tags from a training set. Replicas are then placed in the countries where the highest demand is anticipated.

Key Results

  • Prediction Accuracy: The tag-based approach achieved a median accuracy of 65.9%, significantly outperforming the 33.9% accuracy of a baseline model that assumes videos follow general global traffic patterns.
  • Cache Performance: In simulations using a global storage environment with an LRU (Least Recently Used) cache, the proactive tag-based placement improved global hit rates by 5.6% to 6.8% over random placement.
  • Niche Content Success: For videos with low geographic spread (entropy < 1.5), which represent 40% of the dataset, tags were exceptionally precise in identifying "home" countries.

Discussion and Future Work

While the results are promising, the study acknowledges several areas for refinement. The current model assumes a simplified infrastructure with a datacenter in every country and treats all video files as having equal storage costs. Future research could incorporate more advanced machine learning techniques, such as Random Forests, to better weight "high-signal" tags. Additionally, combining tag data with real-time social media "cascades" could help account for videos that start local but eventually go viral globally.

Summary updated

Abstract

A large portion of today's Internet traffic originates from streaming and video services. Such services rely on a combination of distributed datacenters, powerful content delivery networks (CDN), and multi-level caching. In spite of this infrastructure, storing, indexing, and serving these videos remains adaily engineering challenge that requires increasing efforts on the part of providers and ISPs. In this paper, we explore how the tags attached to videos by users could help improve this infrastructure, and lead to better performance on a global scale. Our analysis shows that tags can be interpreted as markers of a video's geographic diffusion, with some tags strongly linked to well identified geographic areas. Based on our findings, we demonstrate the potential of tags to help predict distribution of a video's views, and present results suggesting that tags canhelp place videos in globally distributed datacenters. We show in particular that even a simplistic approach based on tags can help predict a minimum of 65.9% of a video's views for a majority of videos, and that a simple tag-based placement strategy is able to improve the hit rate of a distributed on-line video service by up to 6.8% globally over a naive random allocation.
BibTeX
@inproceedings{delbruel2016exploring,
  author = {Stéphane Delbruel and Davide Frey and François Taïani},
  title = {{Exploring the Use of Tags for Georeplicated Content Placement}},
  booktitle = {Ic2e},
  year = {2016},
  doi = {10.1109/ic2e.2016.37},
}
Figure