Introduction
Mobile video traffic is projected to account for 75% of all mobile data, presenting a significant challenge for Content Providers (CPs). While increased traffic offers more revenue opportunities through advertisements, it also imposes substantial delivery costs and network congestion. CPs typically employ two strategies to manage this: Data Sponsoring, where the CP pays for user data to encourage consumption, and Edge Caching, where content is stored locally at WiFi hotspots to bypass expensive cellular networks.
Historically, these techniques have been studied in isolation. However, they are deeply coupled: caching decisions determine which requests can be sponsored via low-cost WiFi, while the overall budget constrains both caching and sponsoring activities. This paper proposes a framework for the joint optimization of these strategies to maximize CP revenue under a fixed total budget.
Methodology
The authors formulate the problem as a two-stage decision process:
- Stage I (Long-term): At the beginning of a period (e.g., daily), the CP determines which videos to cache at specific WiFi locations based on estimated demand.
- Stage II (Real-time): As requests arrive, the CP decides whether to sponsor them. Requests for cached content use "WiFi Sponsoring," while others use more expensive "Cellular Sponsoring."
To handle real-time decisions without future knowledge, a Lyapunov-based online algorithm is employed. This algorithm uses a virtual queue to track budget consumption, ensuring long-term budget adherence while maximizing payoff. Additionally, demand is estimated using an i.i.d. model and a Discrete-Time Markov model to exploit temporal correlations in video popularity.
Experimental Setup
The framework was validated using two large-scale datasets from Beijing:
- iQiYi Video Trace: 3 million watching sessions from 190,000 users over two weeks.
- Tencent WiFi Dataset: Locations of over 166,000 free WiFi Access Points.
By mapping video requests to WiFi locations within a 100m range, the authors simulated a realistic edge-caching environment to test the proposed algorithms against greedy and offline optimal baselines.
Key Results and Insights
The study reveals that joint optimization significantly outperforms isolated strategies. Key findings include:
- Substantial Revenue Gains: Jointly optimizing caching and sponsoring increases CP revenue by 30% to 100% compared to sponsoring-only approaches.
- Algorithm Performance: The Lyapunov-based approach achieves approximately 90% of the performance of an "offline optimal" strategy with perfect foresight and outperforms greedy baselines by 10% to 50%.
- Sponsoring Capacity: Within tested budget constraints, the system sponsored roughly 30% of all user requests, effectively boosting ad impressions.
The "budget tug-of-war" is central to these results: while caching requires upfront investment, it reduces the marginal cost of sponsoring. The joint optimization successfully identifies the optimal balance between these competing costs.
Conclusion and Future Work
This research demonstrates the high potential of integrating edge caching with data sponsoring. Future work could explore dynamic caching, where content is updated more frequently than once a day, and more sophisticated models for user behavior, such as how sponsoring might incentivize users to move toward WiFi hotspots specifically to receive sponsored data.
Summary updated