Introduction
As autonomous driving and augmented reality (AR) applications become integral to the Internet of Vehicles (IoV), the demand for real-time, low-latency data access has surged. Traditional cloud-based storage architectures often fail to meet these requirements due to the significant physical distance between centralized servers and mobile users, resulting in excessive data retrieval latency.
To address these challenges, Edge Computing has emerged as a viable solution by positioning computational and storage resources closer to the network edge. However, edge environments introduce several complexities:
- Storage Constraints: Edge nodes possess limited storage capacity compared to cloud servers.
- Data Fragmentation: Information is distributed across numerous nodes, complicating the localization of specific data segments.
- Reliability Issues: Single-point failures can render cached data inaccessible.
- Index Staleness: Conventional filtering techniques, such as Bloom Filters, lack efficient deletion capabilities, leading to "zombie" entries for outdated or unpopular data.
This paper proposes a two-stage optimization model designed to minimize retrieval latency while ensuring high reliability and efficient index management.
Proposed Methodology
The researchers introduced a framework that integrates strategic collaborative caching with advanced query filtering.
Collaborative Caching via Tabu Search
The first stage focuses on the optimal placement of data across edge nodes. Instead of simple popularity-based or random caching, the model employs a Tabu Search algorithm. This metaheuristic approach explores various storage configurations to identify a global optimum that minimizes total transmission distance.
A key feature of this strategy is the "Power of Two" principle: the algorithm identifies the two most suitable storage nodes for high-traffic data segments. This redundancy ensures that every critical piece of data has a backup within the collaborative domain, effectively mitigating the risk of single-point failures and reducing the average distance data must travel. By treating a cluster of edge servers as a unified "collaborative domain," the system allows nodes to share resources and backups seamlessly, transforming isolated silos into a cohesive network.
Query Optimization via Cuckoo Filters
Once data is strategically cached, the next challenge is rapid localization. The model utilizes Cuckoo Filters deployed on edge nodes to streamline the query process.
Unlike standard Bloom Filters, Cuckoo Filters support dynamic deletion operations. This is crucial for edge environments where data popularity fluctuates; when a file is evicted from the cache, its corresponding fingerprint can be immediately removed from the filter. When a vehicle initiates a query, the filter provides a near-instantaneous response regarding data availability, bypassing the need for exhaustive server storage traversals.
Experimental Evaluation
The proposed model was evaluated through extensive simulations using Python 3.6. The simulation environment modeled a collaborative domain of 5 to 30 edge servers processing 100,000 user queries. To ensure realism, data popularity and vehicle distributions followed a Zipf distribution, reflecting real-world patterns where a minority of content generates the majority of traffic. Query tasks were varied between 0.5 and 1.5 Gigacycles to test the system under different workloads.
Key Results
The optimization model was compared against Random Caching (RC) and Partitioned Group Caching (CCS-AGP) strategies.
- Latency Performance: The proposed model achieved a reduction in average data query latency of 13.65% to 24.2% compared to the RC strategy.
- Competitive Advantage: It outperformed the CCS-AGP strategy by 1.89% to 4.16%, demonstrating the benefits of accounting for non-uniform query frequency distributions.
- Scalability: Performance improved as the number of edge nodes increased, as the collaborative domain provided more opportunities to cache data in close proximity to end-users.
Conclusion and Future Directions
The integration of Tabu Search for collaborative caching and Cuckoo Filters for query optimization significantly enhances the efficiency of vehicular edge networks. By addressing storage limitations and the need for dynamic index management, the model provides a robust framework for low-latency data retrieval.
Future research will focus on:
- Real-World Implementation: Deploying the model on large-scale IoT-based smart platforms to validate simulation results.
- Predictive Caching: Incorporating machine learning to predict data popularity trends proactively.
- Filter Optimization: Enhancing the space utilization of Cuckoo Filters to manage even larger volumes of data fingerprints.
Summary updated