Latency estimation for fog-based internet of things

Jianhua Li, Tiehua Zhang, Jiong Jin, Yingying Yang, Dong Yuan, Longxiang Gao

Abstract

As the Internet of Things (IoT) expands into critical domains like autonomous vehicles and real-time surveillance, low latency has become a fundamental requirement. This paper introduces FILE (Fog-based IoT Latency Estimation), a hybrid framework that combines the strengths of distributed and landmark-based network coordinate systems. By dynamically promoting stable fog nodes to act as landmarks, FILE enables IoT devices to predict inter-node latency with over 90% accuracy. Experimental results demonstrate that FILE significantly improves path selection for time-sensitive applications, reducing propagation delay by optimizing routes within the cloud-fog-things continuum.

Introduction

In the evolving landscape of IoT, "low latency" is no longer a luxury but a hard requirement for safety-critical systems. Fog computing addresses this by moving processing to the network edge. However, predicting end-to-end delay between arbitrary nodes in a dynamic, heterogeneous fog environment remains a significant challenge.

Existing Network Coordinate Systems (NCS), which map network delays into virtual geometric distances, typically fall into two categories: landmark-based systems (e.g., GNP) and distributed systems (e.g., Vivaldi). Landmark-based systems are accurate but fragile, relying on fixed infrastructure. Distributed systems are adaptive but struggle with "churn"—the constant joining and leaving of nodes—and lack the management structure necessary for resource-constrained edge devices. FILE bridges this gap by creating a framework that is both stable for accuracy and flexible for dynamic environments.

Methodology

The FILE framework utilizes a two-phase hybrid approach managed by a centralized cloud-based FILE Server.

  1. The Vivaldi Phase (Landmark Generation): Relatively stable nodes, such as storage or application servers, run the Vivaldi algorithm. These nodes treat network links like physical springs; coordinates are adjusted based on whether the "spring" is compressed or stretched by predicted delays. These nodes report their coordinates to the FILE server.
  2. The GNP Phase (Ordinary Node Positioning): When a transient "thing" (e.g., a sensor or vehicle) joins the network, the FILE server assigns it a set of these dynamic landmarks. The node then uses the Simplex Downhill algorithm to calculate its own coordinates relative to those landmarks without needing direct communication with every other node.

This centralized management allows the FILE server to maintain a global view, dynamically selecting the best landmarks and mitigating the positioning workload for low-power devices.

Experimental Setup and Results

The framework was validated using a mix of real-world latency probes (measuring RTT from routers in Sydney and Melbourne to cloud instances) and a simulation testbed. The simulation utilized GNS3 with Cisco 7204 routers and WANem to inject realistic jitter and delay.

Accuracy and Efficiency

FILE achieves high accuracy with fewer resources than traditional methods. Specifically, FILE reaches over 90% prediction accuracy with only 13 landmarks, whereas the standard GNP approach requires 16 landmarks to achieve the same precision.

Figure 4: Prediction Error Rate Comparison showing FILE's superior accuracy with fewer landmarks compared to GNP.

Latency Reduction in Practice

In a case study involving a Real-time Driving-condition Information System (RDIS), the FILE-optimized route demonstrated a massive improvement over traditional cloud routing. While a standard cloud route averaged ~2038ms of delay, the FILE-optimized fog route reduced this to approximately 265ms.

Figure 6: Instantaneous Network Latency Comparison showing the significant RTT reduction achieved by the FILE-optimized fog route compared to cloud and default fog routes.

Conclusion

FILE provides a robust solution for latency estimation in the IoT-Fog ecosystem by treating stable fog nodes as temporary landmarks. This approach provides the structural benefits of landmark-based systems without their inherent fragility. Future work will address convergence speeds in extremely high-mobility scenarios and investigate security measures against "coordinate attacks" where malicious nodes might report false data to disrupt the coordinate space.

Summary updated

Abstract

Low latency is critical for delay-sensitive applications such as video surveillance, live streaming, and online data analytics. Fog computing enables the emergence of the latency-sensitive internet of things (IoT) network to support real-time applications. While the distance between sensing and processing is minimized in the fog network, the cross-fog latency is yet to be determined. In this paper, we study the components of network delays and develop a latency estimation framework for fog-based IoT. The proposed framework, in particular, precisely predicts the end-to-end inter-node delay along the cloud-fog-things continuum. We investigate the benefits and use cases based on latency estimated by the proposed framework. A case study is further conducted to illustrate the validation and advantages, followed by future research directions.
BibTeX
@inproceedings{li2017latency,
  author = {Jianhua Li and Tiehua Zhang and Jiong Jin and Yingying Yang and Dong Yuan and Longxiang Gao},
  title = {{Latency estimation for fog-based internet of things}},
  booktitle = {ITNAC},
  year = {2017},
  doi = {10.1109/atnac.2017.8215403},
}
Figure