Modeling the Delivery of Coded Packets in D2D Mobile Caching Networks

Ameer Ahmed, Hangguan Shan, Aiping HuangIEEE Access 2019 · OpenReview

Abstract

Caching popular files on the mobile nodes have been seen as a promising solution to improve the network performance. In this paper, we analyze the delivery of coded packets of the requested file in a mobile caching network of nodes each with a few pre-cached packets of different files from a large file library. To model the packet delivery process, we develop a two-dimensional Markov chain framework where each request has a deadline requirement. For the delivery of packets, we consider the mobility aspect of caching network where the nodes request the files from other neighbors while moving among different cells and offload the requested files on the device-to-device link. To characterize the network performance, we derive closed-form expressions for file outage probability, the average delay in receiving file, and average throughput to provide some valuable insights on different parameter settings. The simulation results not only verify the accuracy of the analysis but also shed some light on the scenarios where the performance gap between the mobile and fixed caching networks is prominent.

Introduction

Wireless data traffic is increasingly dominated by asynchronous requests for popular content. Caching this content directly on user devices (D2D caching) allows users to retrieve data from nearby peers rather than the central base station, saving bandwidth and energy. As 5G and 6G networks move toward edge-centric architectures, understanding the interplay between node movement, data coding, and delivery deadlines is crucial. This paper provides a rigorous mathematical framework for understanding how mobile devices can share cached content effectively.

Methodology

The researchers employ several key techniques to build their model:

  • Linear Network Coding (LNC): Instead of sharing raw file chunks, nodes store and transmit "coded" packets. A receiver only needs to collect a specific number of linearly independent packets to reconstruct the original file, simplifying the search process in a mobile environment.
  • Reshuffling Mobility Model: To simulate fast mobility, the authors use an i.i.d. (independent and identically distributed) model where nodes are randomly reshuffled into different clusters at the start of every time slot.
Illustration of the node mobility and cluster-based D2D caching model.
  • 2-D Markov Chain Framework: This is the paper's core contribution. One dimension of the chain tracks the number of packets a node has successfully collected, while the second dimension tracks the elapsed time (slots). This allows the authors to define "absorbing states" for both success (file received) and failure (deadline reached).
Timeline of the packet delivery process with scheduling and deadline constraints.

Key Insights and Analysis

The derived closed-form expressions for file outage probability, average delay, and average throughput provide several valuable insights:

  • Mobility as a Search Strategy: In scenarios where nodes have very small caches, mobility is a major advantage. It allows a requesting node to "visit" different clusters and encounter different neighbors, increasing the chance of finding the specific coded packets it needs.
  • The Static Advantage at Scale: Surprisingly, if nodes have large caches or the network is highly reliable, a static network often outperforms a mobile one. In a static setup, scheduling is more predictable; in a mobile setup, a node might move into a "bad" cluster where no one has the needed data, wasting a time slot.
  • The Congestion Paradox: Increasing the total number of nodes in the network can actually degrade individual performance. More nodes lead to more simultaneous requests, which reduces the probability that any single node will be scheduled to receive data in its cluster.
  • Sensitivity to Transmission Errors: Mobile networks are more sensitive to transmission errors. When a transmission fails in a mobile environment, the node may move away before it can successfully retry the packet with the same neighbor.

Results

The theoretical closed-form expressions matched the simulation results almost perfectly across various cache sizes and node densities. Increasing the cache size ($H$) at each node significantly reduces outage probability and delay. For example, increasing cache size from 120 to 420 packets reduced the outage probability from ~0.98 to ~0.91. Additionally, higher Zipf exponents (meaning a few files are much more popular than others) lead to much better performance, as the content reuse rate increases.

Conclusion and Future Work

The authors acknowledge a few simplifying assumptions that could be expanded in future research. The current "reshuffling" model assumes nodes jump instantly between clusters, whereas real-world mobility is more fluid. Furthermore, the model assumes only one D2D link is active per cluster per slot; future work could explore multi-link scheduling with complex interference cancellation. Finally, integrating rate adaptation would make the model more applicable to real-world LTE/5G environments.

Summary updated

Abstract

Caching popular files on the mobile nodes have been seen as a promising solution to improve the network performance. In this paper, we analyze the delivery of coded packets of the requested file in a mobile caching network of nodes each with a few pre-cached packets of different files from a large file library. To model the packet delivery process, we develop a two-dimensional Markov chain framework where each request has a deadline requirement. For the delivery of packets, we consider the mobility aspect of caching network where the nodes request the files from other neighbors while moving among different cells and offload the requested files on the device-to-device link. To characterize the network performance, we derive closed-form expressions for file outage probability, the average delay in receiving file, and average throughput to provide some valuable insights on different parameter settings. The simulation results not only verify the accuracy of the analysis but also shed some light on the scenarios where the performance gap between the mobile and fixed caching networks is prominent.
BibTeX
@inproceedings{ahmed2019modeling,
  author = {Ameer Ahmed and Hangguan Shan and Aiping Huang},
  title = {{Modeling the Delivery of Coded Packets in D2D Mobile Caching Networks}},
  booktitle = {IEEE Access},
  year = {2019},
  url = {https://openreview.net/forum?id=LvI87tO2pUK},
}
Figure