Optimal job partitioning and allocation for vehicular cloud computing

Abstract

With the emergence of advanced and powerful vehicular computing resources, it has become more challenging to meet the demands for vehicular cloud computing. Typically, partitioning and allocating algorithms, such as MapReduce, are used for parallel computing, but partitioning and merging costs of a task are sacrificed as a result. This inherently causes a delay in queueing for the results of the partitioned tasks, which can result in a significant rise in dynamic vehicular cloud computing. Thus, it is crucial to maximize the parallelism of a task's execution across vehicles taking into consideration the dynamic characteristics of the vehicles. This study proposes an optimal job partitioning and allocating algorithm for vehicular cloud computing. To minimize the overall execution time of a job, the proposed algorithm finds the optimal number of tasks among vehicles. Furthermore, a potential capacity distribution model is also presented representing the dynamic characteristics of the vehicles, and finally, the approximated optimal number of partitions is derived by applying the model. The analysis results of this study are demonstrated through extensive evaluations. (C) 2020 Elsevier B.V. All rights reserved.