This is the latest version of this item.

Abstract

The amount of data being generated at the edge of the Internet is rapidly rising as a result of the Internet of Things (IoT). Vehicles themselves are contributing enormously to data generation with their advanced sensor systems. This data contains contextual information; it's temporal and needs to be processed in real-time to be of any value. Transferring this data to the cloud is not feasible due to high cost and latency. This has led to the introduction of edge computing for processing of data close to the source. However, edge servers may not have the computing capacity to process all the data. Future vehicles will have significant computing power, which may be underutilized, and they may have a stake in the processing of the data. This led to the introduction of a new computing paradigm called vehicular cloud (VC), which consists of interconnected vehicles that can share resources and communicate with each other. The VCs may process the data by themselves or in cooperation with edge servers.
Performance modeling of VCs is important, as it will help to determine whether it can provide adequate service to users. It will enable determining appropriate service strategies and the type of jobs that may be served by the VC such that Quality of service (QoS) requirements are met. Job completion time and throughput of VCs are important performance metrics. However, performance modeling of VCs is difficult because of the volatility of resources. As vehicles join and leave the VC, available resources vary in time. Performance evaluation results in the literature are lacking, and available results mostly pertain to stationary VCs formed from parked vehicles. This thesis proposes novel stochastic models for the performance evaluation of vehicular cloud systems that take into account resource volatility, composition of jobs from multiple tasks that can execute concurrently under different service strategies. First, we developed a stochastic model to analyze the job completion time in a VC system deployed on a highway with service interruption. Next, we developed a model to analyze the job completion time in a VC system with a service interruption avoidance strategy. This strategy aims to prevent disruptions in task service by only assigning tasks to vehicles that can complete the tasks’ execution before they leave the VC. In addition to analyzing job completion time, we evaluated the computing capacity of VC systems with a service interruption avoidance strategy, determining the number of jobs a VC system can complete during its lifetime. Finally, we studied the computing capacity of a robotaxi fleet, analyzing the average number of tasks that a robotaxi fleet can serve to completion during a cycle. By developing these models, conducting various analyses, and comparing the numerical results of the analyses to extensive Monte Carlo simulation results, we gained insights into job completion time, computing capacity, and overall performance of VC systems deployed in different contexts.

Available Versions of this Item

• Performance Modeling of Vehicular Clouds Under Different Service Strategies. (deposited 15 Nov 2023 15:33)
  • Performance Modeling of Vehicular Clouds Under Different Service Strategies. (deposited 17 Nov 2023 19:11) [Currently Displayed]