Collaborative Container-based Parked Vehicle Edge Computing Framework for Online Task Offloading
Khoa Nguyen, Steve Drew, Changcheng Huang, Jiayu Zhou
Abstract
As most vehicles spend over 95% of their time in the parking lots, the powerful computing resources of parked vehicles (PVs) are underutilized, that can be considered as available computing nodes to run tasks as well as an extension of the existing infrastructure. In this paper, we propose EdgePV, a collaborative computing paradigm to efficiently improve online heterogeneous task scheduling. To guarantee service reliability, a container orchestration (e.g. Kubernetes) is advocated to be integrated into this proposed architecture due to its notable advanced features such as load-balancing, auto-healing, resource isolation, security, etc,. Kubernetes coordinates PVs to run sufficient numbers of task replicas, providing high service availability against possible failure caused by the mobility of PVs. We investigate how efficient PVs can handle the online computational tasks during peak hours. We also present the dual cost and utility-aware heuristic algorithm, compared with a set of heuristics to solve the problem of task scheduling, that can be devised for replacing the default scheduler in Kubernetes platform. Extensive simulation results show that our proposed design improves the task acceptance ratios and average costs at least 23% and 64%, respectively, at lowest task arrival rate compared to the cooperated cloud-edge architecture. Furthermore, owners of PVs can significantly benefit from incentives received by sharing the idle resources of their PVs.