Full text
| |
| Author(s): |
Joahannes Bruno Dias da Costa
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Computação |
| Defense date: | 2023-09-05 |
| Examining board members: |
Leandro Aparecido Villas;
Fabrício Aguiar Silva;
Lourenço Alves Pereira Júnior;
Juliana Freitag Borin
|
| Advisor: | Leandro Aparecido Villas; Denis Lima do Rosário |
| Abstract | |
The automobile industry has been continuously investing in the modernization of vehicles, improving their communication and data processing capacities. Following this evolution, the Vehicular Edge Computing (VEC) paradigm emerged to provide computing power and storage capability close to vehicular users. In this scenario, vehicles and communication infrastructures can cooperatively attend vehicular services/applications, aggregating their resources and making them available through Vehicular Clouds (VCs). For this availability to happen, the following processes must be carried out: (i) VC Formation, which is the grouping of vehicles and their available computational resources; and (ii) Task Scheduling, which aims to decide which of the VCs a given set of tasks will be processed. Also, carrying out load balancing between the VCs is essential to increase fairness in the use of resources and make the load distribution in the network more homogeneous. However, mobility is one of the main challenges in proposing solutions in these scenarios, since vehicular mobility causes several changes in the network topology and intermittent connections. In this context, this thesis presents a study of how VCs are formed and how applications can use the resources of these clouds efficiently for data processing and, with that, help in decision making that requires low latency and restricted processing time. Furthermore, this thesis proposes a series of mechanisms to deal with different aspects of mobility at the edge of the vehicular network. The first contribution of this thesis lies in a mobility-aware solution to estimate the dwell-time of vehicles in a given region and thus mitigate the impacts of mobility on the VC formation process. Secondly, a task scheduling mechanism was proposed that uses a Recurrent Neural Network (RNN) architecture to estimate computational resources in VCs and ensure that user demands are met. This approach manages to increase the number of scheduled tasks, decrease the overall system latency, and reduce the monetary costs for using computational resources. The third contribution focuses on increasing fairness and load balancing in the use of VC’s resources. The main advantages of the latter mechanism include: (i) a task scheduler that maximizes the number of tasks successfully scheduled and processed while maintaining fair load balancing in the use of computational resources and (ii) the use of multithreading for parallel solving of scheduling subproblems, aiming to reduce system latency without compromising the overall performance of the solution. The proposed solutions were widely compared with other state-of-the-art solutions in different performance evaluation metrics and considering realistic mobility scenarios. The results show that the proposed approaches are efficient, scalable, and cost-effective, which can be good alternatives to mitigate the challenges imposed by the dynamics of vehicular mobility in the VEC environments (AU) | |
| FAPESP's process: | 18/16703-4 - Vehicular cloud computing for information management in intelligent transportation systems |
| Grantee: | Joahannes Bruno Dias da Costa |
| Support Opportunities: | Scholarships in Brazil - Doctorate |