Joint pushing and caching with a finite receiver buffer: Optimal policies and throughput analysis

Wei Chen, H. Vincent Poor

Abstract

This paper investigates the Joint Pushing and Caching (JPC) problem in content-centric wireless networks where receivers have finite storage capacity. By leveraging Request Delay Information (RDI)—the time elapsed between content generation and user request—the authors develop optimal offline and online policies to maximize effective throughput. The research demonstrates that incorporating causal feedback about past requests can significantly enhance performance, effectively doubling throughput in scenarios with limited buffer sizes.

Introduction

In modern wireless networks, proactive content delivery ("pushing") and local storage ("caching") are essential strategies for mitigating peak-hour congestion. However, most existing literature assumes infinite storage or focuses solely on content selection. This paper identifies a critical tension in finite-buffer systems: pushing content too early may cause buffer overflows, while pushing too late may result in the content failing to arrive before the user's request deadline. To address this, the authors propose a JPC framework that jointly optimizes transmission timing and buffer eviction policies.

System Model and Methodology

The authors frame the JPC problem using a discrete-time model where a Base Station (BS) pushes content to a user with a buffer of size $N$. The core of their approach is the use of Request Delay Information (RDI), a random variable representing the time until a specific file is requested.

Figure 1: System model illustrating the timeline of content pushing and user requests relative to Request Delay Information (RDI).

Three primary policies are developed based on the availability of RDI:

  1. Offline JPC (Noncausal RDI): Assumes perfect knowledge of future request times, providing a theoretical performance upper bound using network calculus.
  2. Online JPC (Statistical RDI): Relies on the probability distribution of requests. The authors employ trellis-aided dynamic programming to navigate possible buffer states and maximize expected throughput with manageable computational complexity.
  3. Online JPC (Causal RDI): Enhances the statistical approach by incorporating real-time feedback. Once a user requests a file, the BS is notified and can immediately evict that file from the buffer, freeing space for new content.

Key Results and Insights

The proposed policies were validated through extensive numerical simulations and theoretical derivations. A primary finding is the Buffer Bottleneck: for systems relying only on statistical information, throughput is strictly constrained by the physical buffer size, regardless of the pushing channel's capacity.

Figure 3: Throughput performance comparison of Noncausal, Statistical, and Causal RDI policies as a function of buffer size N.

Key performance highlights include:

  • The Power of Feedback: Causal RDI provides over a 100% throughput gain compared to purely statistical methods when the buffer size is small (e.g., $N < 4$). This feedback allows a small buffer to emulate the performance of a much larger one by efficiently removing "dead weight."
  • Asymptotic Behavior: In the offline case, as the pushing duration ($K$) increases, throughput approaches the total number of files ($L$), following the derived approximation $R \approx NK(1 - e^{-L/NK})$.
  • Theoretical Validation: The authors' closed-form analytical approximations for throughput (Equations 18 and 26 in the full paper) show a near-perfect match with simulation results, offering a robust tool for network planning.

Conclusion and Future Work

This work establishes that joint optimization of pushing and caching, supported by causal request feedback, is vital for maximizing the efficiency of finite-buffer wireless systems. While the current study assumes independent and identically distributed (i.i.d.) request delays and a single-user link, future research will explore more complex scenarios. These include multicast delivery to multiple users with diverse RDI profiles and the impact of correlated content popularity on optimal eviction strategies.

Summary updated

BibTeX
@inproceedings{chen2016joint,
  author = {Wei Chen and H. Vincent Poor},
  title = {{Joint pushing and caching with a finite receiver buffer: Optimal policies and throughput analysis}},
  booktitle = {ICC},
  year = {2016},
  doi = {10.1109/icc.2016.7510697},
}
Figure