Online Edge Caching and Wireless Delivery in Fog-Aided Networks with Dynamic Content Popularity

Seyyed Mohammadreza Azimi, Osvaldo Simeone, Avik Sengupta, Ravi Tandon

Abstract

Fog Radio Access Network (F-RAN) architectures can leverage both cloud processing and edge caching for content delivery to the users. To this end, F-RAN utilizes caches at the edge nodes (ENs) and fronthaul links connecting a cloud processor to ENs. Assuming time-invariant content popularity, existing information-theoretic analyses of content delivery in F-RANs rely on offline caching with separate content placement and delivery phases. In contrast, this work focuses on the scenario in which the set of popular content is time-varying, hence necessitating the online replenishment of the ENs' caches along with the delivery of the requested files. The analysis is centered on the characterization of the long-term Normalized Delivery Time (NDT), which captures the temporal dependence of the coding latencies accrued across multiple time slots in the high signal-to-noise ratio regime. Online edge caching and delivery schemes are investigated for both serial and pipelined transmission modes across fronthaul and edge segments. Analytical results demonstrate that, in the presence of a time-varying content popularity, the rate of fronthaul links sets a fundamental limit to the long-term NDT of F-RAN system. Analytical results are further verified by numerical simulation, yielding important design insights.

Introduction: The Challenge of Dynamic Popularity

In the evolution of 5G and beyond, Fog Radio Access Networks (F-RANs) have emerged as a promising architecture to minimize latency by bringing content closer to the end-user. By deploying Edge Nodes (ENs) equipped with local storage and processing capabilities, the network can serve popular requests—such as viral videos—without relying solely on the high-latency backhaul to a central cloud.

Traditionally, research in this field has focused on offline caching, where content is placed in the edge caches during off-peak hours based on static popularity assumptions. However, real-world content popularity is highly volatile; a specific file may become viral and then obsolete within hours. This research addresses the gap by modeling online edge caching, where the set of popular files $\mathcal{F}_t$ evolves over time, requiring the system to simultaneously deliver requested content and replenish its caches over a limited-capacity fronthaul link.

Figure 1: F-RAN architecture illustrating the dynamic content set $\mathcal{F}_t$ and the delivery process across cloud, edge nodes, and users.

Methodology and System Model

The study employs an information-theoretic framework to evaluate performance in the high signal-to-noise ratio (SNR) regime. The primary metric used is the Normalized Delivery Time (NDT), which represents the ratio of the delivery time in the F-RAN system to the time required in an ideal, interference-free system with local content availability.

Content Churn and Transmission Modes

The authors model content dynamics using a Markov process: in each time slot, a new file enters the popular set with probability $p$, replacing an existing file. To manage this churn, the system operates in two primary modes:

  1. Serial Transmission: The fronthaul transmission from the cloud to the ENs occurs first, followed by the wireless delivery from the ENs to the users.
  2. Pipelined Transmission: The fronthaul and wireless links operate simultaneously, allowing for potential latency overlaps.
Figure 2: Temporal structure of a time slot, showing the allocation of time for fronthaul ($T_F$) and edge ($T_E$) transmissions.

Caching Strategies

Several strategies are explored to optimize the long-term NDT:

  • Reactive vs. Proactive: Reactive schemes update the cache only upon a user request for a missing file, while proactive schemes push new popular files to the edge before they are requested.
  • Adaptive Fractional Caching: A novel approach where the system may choose to cache only a fraction of a file, even if storage space is available, to conserve fronthaul bandwidth when popularity changes rapidly.

Key Insights: The Fronthaul Bottleneck

The analysis reveals that in dynamic environments, the fronthaul link rate is the fundamental bottleneck. While large edge caches are beneficial in static scenarios, their utility is capped by the speed at which they can be updated. The research quantifies this through the "latency gap," showing that the long-term NDT is the sum of the static (offline) NDT and an additive penalty inversely proportional to the fronthaul rate.

Furthermore, the study demonstrates that pipelining offers significant advantages. By utilizing the fronthaul "for free" during the wireless delivery phase of other content, pipelined systems can effectively hide the latency associated with cache replenishment. In contrast, serial systems suffer more acutely from content churn.

Results and Design Implications

  • Mathematical Bounds: The paper establishes that the minimum long-term NDT $\bar{\delta}^*_{on}$ is approximately $c\delta^*_{off} + O(1/r)$, where $r$ is the fronthaul rate. This confirms that the popularity penalty is intrinsically linked to fronthaul capacity.
  • Eviction Policies: Simulations indicate that when the popular set is not perfectly known, Least Recently Used (LRU) and First-In-First-Out (FIFO) strategies are highly effective, closely approaching the performance of systems with perfect knowledge of popularity.
  • Adaptive Caching Benefits: In scenarios with high churn, caching less than the maximum capacity can prevent the fronthaul from becoming overwhelmed by updates that may never be served before the content becomes unpopular.

Conclusion and Future Directions

This work provides a rigorous foundation for understanding online caching in fog networks. It highlights that network designers must balance cache size with fronthaul capacity to handle dynamic content effectively. Future research could extend these models to include non-uniform request distributions (e.g., Zipf distributions), account for imperfect Channel State Information (CSI), and explore multi-tier cloud-to-edge hierarchies.

Summary updated

Abstract

Fog Radio Access Network (F-RAN) architectures can leverage both cloud processing and edge caching for content delivery to the users. To this end, F-RAN utilizes caches at the edge nodes (ENs) and fronthaul links connecting a cloud processor to ENs. Assuming time-invariant content popularity, existing information-theoretic analyses of content delivery in F-RANs rely on offline caching with separate content placement and delivery phases. In contrast, this work focuses on the scenario in which the set of popular content is time-varying, hence necessitating the online replenishment of the ENs' caches along with the delivery of the requested files. The analysis is centered on the characterization of the long-term Normalized Delivery Time (NDT), which captures the temporal dependence of the coding latencies accrued across multiple time slots in the high signal-to-noise ratio regime. Online edge caching and delivery schemes are investigated for both serial and pipelined transmission modes across fronthaul and edge segments. Analytical results demonstrate that, in the presence of a time-varying content popularity, the rate of fronthaul links sets a fundamental limit to the long-term NDT of F- RAN system. Analytical results are further verified by numerical simulation, yielding important design insights.
BibTeX
@article{azimi2018online,
  author = {Seyyed Mohammadreza Azimi and Osvaldo Simeone and Avik Sengupta and Ravi Tandon},
  title = {{Online Edge Caching and Wireless Delivery in Fog-Aided Networks with Dynamic Content Popularity}},
  journal = {IEEE J. Sel. Areas Commun.},
  year = {2018},
  url = {https://openreview.net/forum?id=m0cyZFGq04U},
}
Figure