Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On-Demand Video Dispatch Networks: A Scalable End-to-End Learning Approach

Damao Yang, Sihan Peng, He Huang, Hongliang Xue

TL;DR

The paper tackles scalable, peak-aware video-on-demand dispatch by proposing two coupled neural networks: a clustering network to group videos by request patterns and a policy network to assign dispatch probabilities to video clusters. Both networks share temporal feature extractors built from a CNN for intraday patterns and an RNN for interday dynamics, enabling end-to-end training with MBGD. The approach achieves substantial improvements over baselines in real-world data, including significant gains in peak-time prediction accuracy and dispatch efficiency, aided by asynchronous training and a post-processing stage that maps cluster-level predictions to CDN-level assignments. This design demonstrates scalable handling of billions of videos and variable peak/off-peak demands, with practical significance for reducing peak-hour bandwidth use while maintaining or improving peak-time quality.

Abstract

We design a dispatch system to improve the peak service quality of video on demand (VOD). Our system predicts the hot videos during the peak hours of the next day based on the historical requests, and dispatches to the content delivery networks (CDNs) at the previous off-peak time. In order to scale to billions of videos, we build the system with two neural networks, one for video clustering and the other for dispatch policy developing. The clustering network employs autoencoder layers and reduces the video number to a fixed value. The policy network employs fully connected layers and ranks the clustered videos with dispatch probabilities. The two networks are coupled with weight-sharing temporal layers, which analyze the video request sequences with convolutional and recurrent modules. Therefore, the clustering and dispatch tasks are trained in an end-to-end mechanism. The real-world results show that our approach achieves an average prediction accuracy of 17%, compared with 3% from the present baseline method, for the same amount of dispatches.

On-Demand Video Dispatch Networks: A Scalable End-to-End Learning Approach

TL;DR

The paper tackles scalable, peak-aware video-on-demand dispatch by proposing two coupled neural networks: a clustering network to group videos by request patterns and a policy network to assign dispatch probabilities to video clusters. Both networks share temporal feature extractors built from a CNN for intraday patterns and an RNN for interday dynamics, enabling end-to-end training with MBGD. The approach achieves substantial improvements over baselines in real-world data, including significant gains in peak-time prediction accuracy and dispatch efficiency, aided by asynchronous training and a post-processing stage that maps cluster-level predictions to CDN-level assignments. This design demonstrates scalable handling of billions of videos and variable peak/off-peak demands, with practical significance for reducing peak-hour bandwidth use while maintaining or improving peak-time quality.

Abstract

We design a dispatch system to improve the peak service quality of video on demand (VOD). Our system predicts the hot videos during the peak hours of the next day based on the historical requests, and dispatches to the content delivery networks (CDNs) at the previous off-peak time. In order to scale to billions of videos, we build the system with two neural networks, one for video clustering and the other for dispatch policy developing. The clustering network employs autoencoder layers and reduces the video number to a fixed value. The policy network employs fully connected layers and ranks the clustered videos with dispatch probabilities. The two networks are coupled with weight-sharing temporal layers, which analyze the video request sequences with convolutional and recurrent modules. Therefore, the clustering and dispatch tasks are trained in an end-to-end mechanism. The real-world results show that our approach achieves an average prediction accuracy of 17%, compared with 3% from the present baseline method, for the same amount of dispatches.

Paper Structure

This paper contains 24 sections, 19 equations, 11 figures, 3 tables.

Table of Contents

  1. Introduction and Related Work
  2. Time Series Forecasting
  3. Ranking Problems
  4. Clustering and Manifold Learning
  5. End-to-end Learning
  6. Our Contributions
  7. Notations
  8. Problem Formulation
  9. The Structure of Dispatch Neural Networks
  10. Primitive Inputs
  11. Temporal Layers
  12. Clustering Layers
  13. Accumulation Layers
  14. Policy Layers
  15. Implementation and Results
  16. ...and 9 more sections

Figures (11)

  • Figure 1: Video Dispatch System and Problems
  • Figure 2: Structure of the Two Coupled Networks
  • Figure 3: Temporal layers
  • Figure 4: Example of Hierarchical Clustering Blocks
  • Figure 5: Peak-time Video Request Distribution
  • ...and 6 more figures