Wireless Transmission of Big Data: Data-Oriented Performance Limits and Their Applications

Hong-Chuan Yang, Mohamed-Slim AlouiniarXiv 2018 · arXiv

Introduction

Traditional wireless system design has long been anchored in channel-oriented metrics, primarily ergodic capacity and outage capacity. While effective for legacy services like voice and continuous video streaming, these metrics are increasingly inadequate for the "Big Data" and Internet of Things (IoT) era. Modern data traffic is characterized by bursty sessions where the objective is to transmit a specific file size ($H$) within a limited number of channel coherence times. In such scenarios, long-term averages are less relevant than the probability of a specific session meeting its deadline. This paper proposes a paradigm shift toward "data-oriented" performance limits to better characterize these modern requirements.

Data-Oriented Performance Metrics

To address the limitations of traditional metrics, the authors move away from standard capacity formulas to define two primary data-oriented metrics:

  1. Minimum Transmission Time (MTT) and Delay Outage Rate (DOR): MTT represents the absolute minimum time required to deliver a fixed data volume $H$ over a fading channel. The DOR is the probability that the actual transmission time exceeds a predefined delay threshold.
  2. Maximum Entropy Throughput (MET) and Information Outage Rate (IOR): MET quantifies the maximum information deliverable within a fixed time-frequency resource block. The IOR is the probability that the system fails to meet a target data volume within that timeframe.

These metrics allow for session-specific optimization that accounts for the bursty nature of Big Data and IoT traffic, focusing on the reliability of individual transmissions rather than long-term averages.

Comparative Analysis of Transmission Strategies

The paper re-evaluates two fundamental transmission strategies using these new metrics:

  • Optimal Rate Adaptation (ORA): Maintains constant power while adapting the data rate to match instantaneous channel conditions.
  • Optimal Power and Rate Adaptation (OPRA): Employs "water-filling" logic, adapting both power and rate to maximize average throughput by saving power during deep fades.

While OPRA is traditionally considered superior for maximizing ergodic capacity, the data-oriented analysis reveals a different story for session-based reliability.

Key Results and Discussion

The analysis demonstrates that the choice of strategy depends heavily on the specific session requirements. For instance, at low average signal-to-noise ratios (SNR) or when the allowed delay is relatively large, ORA can surprisingly outperform OPRA.

Delay Outage Rate (DOR) comparison between ORA and OPRA for different data sizes (H).

As shown in Figure 1, for very small delay thresholds, OPRA (adaptive power) provides a lower outage rate. However, as the delay threshold increases, ORA (constant power) becomes significantly more reliable. A critical observation is the "outage floor" seen in OPRA; because OPRA may choose not to transmit at all during poor channel conditions to save power for future "better" slots, it can fail to meet a deadline that a persistent strategy like ORA would satisfy. This highlights the "averages trap" where maximizing long-term efficiency compromises short-term reliability.

Conclusion

The shift to data-oriented metrics reveals that traditional "optimal" strategies like water-filling are not universally best for Big Data and IoT applications. The optimal transmission strategy is session-dependent, varying with file size, delay constraints, and SNR. Future work will involve extending these models to scenarios with imperfect channel state information (CSIT) and more complex multi-user network topologies.

Summary updated

Abstract

The growing popularity of big data and Internet of Things (IoT) applications bring new challenges to the wireless communication community. Wireless transmission systems should more efficiently support the large amount of data traffics from diverse types of information sources. In this article, we introduce a novel data-oriented approach for the design and optimization of wireless transmission strategies. Specifically, we define new performance metrics for individual data transmission session and apply them to compare two popular channel-adaptive transmission strategies. We develop several interesting and somewhat counterintuitive observations on these transmission strategies, which would not be possible with conventional approach. We also present several interesting future research directions that are worth pursuing with the data-oriented approach.
BibTeX
@misc{yang2018wireless,
  author = {Hong-Chuan Yang and Mohamed-Slim Alouini},
  title = {{Wireless Transmission of Big Data: Data-Oriented Performance Limits and Their Applications}},
  year = {2018},
  eprint = {1805.09923},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1805.09923},
}
Figure