Task scheduling in fog computing systems using deep learning

Abstract

Optimal resource allocation is one of the fundamental challenges in fog computing systems. In this paper, a novel deep learning method is proposed for efficient task allocation and scheduling in fog computing systems. The proposed method consists of a three-phase structure. In the first phase, a stable communication structure between network components is established using a minimum spanning tree structure based on the distance between nodes. In the second phase, a deep neural network is used to determine the appropriate computational resources for each task, in which task characteristics and processing resources are analysed as inputs and the priority of assigning the task to each resource is determined. And finally, in the third phase, after assigning each task to resources, the order and processing time of the tasks are determined separately using an improved version of the round-robin algorithm, and this is done based on the dependencies between tasks. To check our method, we performed experiments by changing the number of running tasks and by altering the average time needed to complete each task. The method is measured against well-known algorithms such as enhanced round-robin, particle swarm optimization and deep reinforcement learning for multi-objective scheduling. The experiments conducted and the values obtained in terms of response time, turnaround time and waiting time show that the proposed method has achieved 9.96 %, 7.80 % and 8.69 % improvements in the first scenario, respectively. Moreover, in the second scenario, these successes have increased to 20.25 %, 8.94 %, and 7.33 %, respectively. These results represent the high efficiency of the proposed method for the optimization of different times and performance improvement of a system under various conditions.