Social-aware device-to-device communication: a contribution for edge and fog computing?

Huber Flores, Denzil Ferreira, Chu Luo, Vassilis Kostakos, Pan Hui, Rajesh Sharma, Sasu Tarkoma, Yong LiUbiComp Adjunct 2016 · DOI

TL;DR

This paper explores the practical viability of Device-to-Device (D2D) communication for edge and fog computing by developing and deploying a "Detector" sensor. Their real-world study reveals that mobile users are consistently in proximity to other devices, suggesting significant potential for coordinated task processing and offloading, rather than relying solely on individual device capabilities.

What Problem Does This Paper Address?

The paper tackles the question of whether the theoretical benefits of Device-to-Device (D2D) communication can be practically realized in real-world scenarios, particularly in the context of emerging paradigms like edge and fog computing. While D2D promises to augment mobile device capabilities and enhance user experience by enabling opportunistic infrastructure formation, its actual utility and the extent to which it can be leveraged remain unclear. The authors highlight that existing research often focuses on theoretical potential or simulations, leaving a gap in understanding how D2D performs "in the wild." This is crucial because edge and fog computing rely heavily on mobile devices to support processing for other devices (like sensors), and D2D is a key enabler for this.

Methods

The core contribution is the development of a software sensor called Detector. This application runs in the background on mobile devices and actively searches for other discoverable devices using Bluetooth and Wi-Fi Direct. The discovery process is intentionally throttled to occur every 15 minutes to minimize battery drain. The Detector application collects data on discovered devices, including their type, name, and communication interface addresses, storing this information in a local SQLite database. This data is then uploaded daily to a remote server for analysis. The authors also describe the setup and methodology for their experiment, which involved recruiting 15 participants to install the Detector application and collecting data over a one-month period.

Datasets

The paper introduces a newly collected dataset from 15 participants who installed the "Detector" application over a one-month period. This dataset comprises traces of discovered devices via Bluetooth and Wi-Fi Direct, along with information about battery levels and usage patterns. The authors note that data from 5 participants was discarded due to insufficient information for analysis, leaving a dataset from 10 participants. This dataset is made publicly available on GitHub.

Key Insights and Ideas

  • Ubiquitous Proximity: The most striking finding is that, on average, a mobile user is in proximity to at least one other mobile device throughout the day. This strongly supports the idea that devices can indeed form opportunistic D2D infrastructure.
  • Practicality of D2D: The study moves beyond simulations to demonstrate the real-world presence of D2D infrastructure, suggesting that the theoretical benefits of D2D for edge and fog computing are not just hypothetical.
  • Task Scheduling Potential: The consistent presence of nearby devices implies that mobile devices can schedule task processing and offloading in coordination with others, potentially leveraging more powerful devices, rather than handling all processing independently.
  • Battery Drain is a Concern: The discovery process, while essential, consumes significant battery life (approximately 12 hours of battery life for detection). This highlights the need for energy-efficient discovery mechanisms and careful scheduling.
  • Wi-Fi Direct Dominance: Wi-Fi Direct appears to be a more effective communication interface for discovering a larger number of addressable devices compared to Bluetooth.

Key Results

  • Consistent Discoverability: The study found that mobile users are, on average, co-located in proximity to at least one other mobile device throughout the day.
  • Participant Variability: The number of devices detected varied significantly among participants, reflecting differences in their daily routines and mobility patterns. Some participants discovered a large number of devices, while others discovered fewer.
  • Battery Consumption: The detection process using the Detector application consumes approximately 12 hours of battery life.
  • Wi-Fi Direct vs. Bluetooth: Wi-Fi Direct generally provided a larger number of discoverable devices compared to Bluetooth.
  • User Energy Habits: 80% of users maintained their mobile devices above 60% battery charge, suggesting a preference for higher energy levels when considering resource leasing for D2D interactions.

Limitations and Future Work

  • Discoverability Focus: The study primarily focuses on the "discoverable" aspect of D2D infrastructure formation, leaving the establishment of stable inter-communication channels and the coordination of devices for processing load as areas for future work.
  • Battery Drain Mitigation: While the discovery interval was set to 15 minutes to reduce battery drain, the paper acknowledges that this process still consumes significant energy, indicating a need for more energy-efficient discovery mechanisms.
  • Stability Measurement: Measuring the stability of D2D infrastructure is challenging due to privacy concerns and varying device visibility (e.g., Bluetooth/Wi-Fi being turned off). The paper suggests that future work could model stability based on detection frequency and duration.
  • Community Identification: The authors plan to investigate the identification of stable communities of devices that can be trusted for analytical roles in edge computing.
  • Hierarchical Coordination: Future work will explore hierarchical dissemination approaches for task propagation to overcome communication interface limitations in distributing computational tasks.
  • Cross-Device Sensing and Crowdsourcing: The paper suggests that D2D infrastructure can be leveraged for cross-device sensing (creating "big sensors") and for social-incentivized crowdsourcing.

Summary updated

BibTeX
@inproceedings{flores2016social,
  author = {Huber Flores and Denzil Ferreira and Chu Luo and Vassilis Kostakos and Pan Hui and Rajesh Sharma and Sasu Tarkoma and Yong Li},
  title = {{Social-aware device-to-device communication: a contribution for edge and fog computing?}},
  booktitle = {UbiComp Adjunct},
  year = {2016},
  doi = {10.1145/2968219.2968589},
}
Figure