Abstract
The rapid expansion of smart cities relies on Wireless Sensor Networks (WSNs) that often face significant resource constraints. While edge computing offers a solution to high-latency cloud offloading, deploying edge nodes at scale introduces challenges in reliability and environmental sustainability. This paper presents ARES (sustAinable and Resilient Edge proviSioning), a two-stage optimization framework that balances the "iron triangle" of low latency, high reliability, and energy efficiency. By integrating multi-objective genetic algorithms with Dynamic Bayesian Networks to model correlated failures, ARES provides a robust methodology for urban IoT deployment.
Introduction: The Edge Computing Dilemma
Wireless Sensor Networks are the sensory organs of modern smart cities, yet they are inherently limited by low processing power and finite energy reserves. Traditional cloud offloading introduces latencies—averaging around 74ms—that are unacceptable for safety-critical applications like autonomous traffic management.
Edge computing addresses this by moving computation closer to the data source. However, edge nodes (ENs) are typically less reliable than centralized data centers. To compensate, providers often resort to over-provisioning, which leads to excessive energy consumption and a significant carbon footprint. The core challenge lies in finding a "Goldilocks" deployment: a configuration that ensures high reliability and low latency without compromising sustainability.
The ARES Framework
The ARES framework employs a two-stage optimization process to navigate these conflicting objectives.
Stage 1: Efficiency Optimization
Using the NSGA-II (Non-dominated Sorting Genetic Algorithm II), ARES first identifies a "Pareto front" of candidate solutions. These solutions represent the optimal trade-offs between transmission time and energy consumption, where one cannot be improved without degrading the other.
Stage 2: Reliability Augmentation
In the second stage, ARES evaluates these candidate solutions against a specific Reliability Service Level Agreement (SLA). If a solution fails to meet the reliability threshold, the framework "augments" it by exploring neighboring cells to find more resilient node placements.
A critical innovation in ARES is the use of Dynamic Bayesian Networks (DBN) to model system reliability. Unlike standard models that assume independent node failures, ARES accounts for correlated failures—such as localized power outages—providing a more realistic estimation of system-wide uptime in dense urban environments.
Experimental Evaluation
The researchers validated ARES using a combination of real-world datasets and large-scale simulations.
- Vienna Smart Traffic Lights: A dataset comprising 1,369 traffic light locations in Vienna, Austria, was used to simulate a wide-area WSN.
- SETI@home Failure Traces: To model the high "churn" and variable reliability of edge environments, the study utilized failure data from over 226,000 personal computers.
- InTraSafEd5G: A real-world traffic safety use case (blind-spot detection) served as the primary workload model for performance testing.
Key Insights and Results
The evaluation of ARES yielded several critical findings for the future of urban IoT:
- The Impact of Correlated Failures: Ignoring the spatial and temporal dependencies of node failures leads to a significant overestimation of reliability. ARES’s DBN-based approach ensures that deployments are not "fragile" to localized events.
- Efficiency and Reliability are Compatible: ARES demonstrated that "five-nines" (99.999%) availability can be achieved with only a marginal increase in energy consumption (~1.47%) compared to the most energy-efficient, yet unreliable, configurations.
- Performance Gains: Compared to standard three-objective optimization (3OBJ) baselines, ARES reduced system unavailability by 40%. Furthermore, even when optimized for maximum reliability, the mean transmission time remained within 9.27% of the absolute shortest possible time.
Conclusion and Future Work
ARES provides a computationally tractable solution for provisioning reliable and sustainable edge infrastructure in smart cities. By decoupling efficiency optimization from reliability hardening, it manages complex trade-offs that were previously difficult to balance.
Future work will focus on transitioning ARES from an offline provisioning tool to an "online" system capable of real-time re-optimization in response to dynamic environmental conditions and bursty traffic patterns.
Summary updated