Facing serious challenges in bandwidth and delay, currently adopted cloud computing is no longer effective for providing real-time Internet of Vehicles (IoV) services in the intelligent transportation system (ITS). A newly introduced computing paradigm with a distributed feature, i.e., edge computing, can be a complement to the centralized cloud computing with computation offloaded to the distributed edge servers (ESs). Typically, edge servers are the host of IoV services in edge computing, which requires an appropriate quantification and placement before the implementation of computation offloading. To pursue a higher quality of service (QoS), the quantity and locations of the ESs need to be thoroughly discussed ahead. Otherwise, additional delay and network congestion will occur. Simultaneously, as the ITS is continuously developing, the existing placement of ESs ought to be adjusted to be in line with the dynamic change of IoV traffic. However, ES placement schemes are often devised by clustering methods with a fixed ES quantity and are unaware of the extensibility of the placement. To address the problems that mentioned, a dynamic ES placement approach (DEP) is developed. Technically, DEP leverages the non-dominated sorting genetic algorithm III (NSGA-III) for placements with better performance and less reconstruction of existing placement. The population of NSGA-III is initialized with clustering algorithms for a higher accuracy and convergence speed, and the fitness of minimum reconstruction cost is calculated based on Kuhn–Munkres bipartite graph matching algorithm. A real-world traffic data with 436 deployed roadside units in the provincial capital city Nanjing, China, is leveraged for comparative experiments. The experimental results verify the effectiveness of DEP with a 17.64% lower latency and 25.82% lower workload standard deviation comparing to the clustering method.
- Bipartite graph matching
- Edge server placement
- Evolutionary multi-objective optimization
- Internet of Vehicles