TPAoI: Ensuring Fresh Service Status at the Network Edge in Compute-First Networking

Haosheng He, Jianpeng Qi, Chao Liu, Junyu Dong, Yanwei YuarXiv 2024 · DOI · arXiv

Abstract

In edge computing and Internet of Things (IoT) environments, data freshness is critical for time-sensitive applications. Traditional Age of Information (AoI) metrics often overlook the complete service loop, leading to suboptimal network performance. This paper proposes Three-Phase Age of Information (TPAoI), a holistic metric that accounts for updating, accessing, and requesting delays. Using a Dueling Double Deep Q-Network (D3QN), we demonstrate a 47% reduction in information staleness and a 48% decrease in unnecessary network traffic compared to existing standards.

Introduction

The rapid growth of Edge Computing has made "freshness" a primary requirement for status updates in systems like autonomous vehicles and remote medical monitoring. While Age of Information (AoI) is the standard metric for measuring data staleness, current models frequently ignore the "integrity" of the service loop. Specifically, they fail to integrate three critical stochastic phases:

  1. Updating: Server-to-Access Point (AP) push time.
  2. Accessing: The random timing of user data checks.
  3. Requesting: The delay in sending requests back to the remote service.
The three stochastic phases of the TPAoI model: Updating, Accessing, and Requesting.

Ignoring these delays results in updates that are either too late to be useful or too frequent, causing network congestion. This research bridges the gap by modeling the entire lifecycle of a service request.

Methodology

The researchers developed the TPAoI metric to mathematically aggregate delays across all three phases. To optimize this in dynamic environments, they employed Reinforcement Learning:

  • Markov Decision Process (MDP): The system state is modeled using five sub-states: current AoI, transmission timers, and user access intervals.
  • D3QN Algorithm: A Dueling Double Deep Q-Network was used to avoid overestimation bias and better distinguish between state values and action advantages (e.g., deciding whether to update now or wait).
  • Penalty Mechanism: A cost-based reward function penalizes excessive updates, forcing the agent to learn the most efficient update frequency.

Results and Discussion

Experimental evaluations were conducted in a simulation environment mimicking dynamic edge conditions with Poisson-distributed user access and varied network jitters (Exponential and Normal distributions).

Performance comparison of TPAoI against QAoI and Conventional AoI under different transmission delay distributions.

Key findings include:

  • Enhanced Freshness: TPAoI achieved a 47% reduction in average AoI compared to Query AoI (QAoI) benchmarks.
  • Operational Efficiency: The model reduced status update frequency by 48% compared to conventional AoI models, demonstrating that "smarter" updates are more effective than "more" updates.
  • Robustness: The D3QN agent showed stable convergence and maintained high performance across different network delay distributions (Exponential vs. Normal).
  • Predictive Capability: The agent learned to anticipate user access, triggering updates just in time to account for transmission delays.

Conclusion and Future Work

TPAoI provides a more comprehensive framework for managing data freshness at the network edge. By accounting for the full request-response cycle, it significantly improves both service quality and network efficiency. Future research will focus on scaling this framework to complex mesh topologies involving multiple edge servers and access points, and validating the model against real-world network traffic traces.

Summary updated

BibTeX
@misc{he2024tpaoi,
  author = {Haosheng He and Jianpeng Qi and Chao Liu and Junyu Dong and Yanwei Yu},
  title = {{TPAoI: Ensuring Fresh Service Status at the Network Edge in Compute-First Networking}},
  year = {2024},
  eprint = {2412.18391},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2412.18391},
}
Figure