A reinforcement learning-based load balancing algorithm for fog computing

Fog computing is a developing paradigm for bringing cloud computing capabilities closer to end-users. Fog computing plays an important role in improving resource utilization and decreasing delay for internet of things (IoT) applications. At the same time, it faces many challenges, including challenges related to energy consumption, scheduling and resource overload. Load balancing helps to reduce delay, increase user satisfaction, and also increase system efficiency by efficiently and fairly allocation of tasks among computing resources. Fair load distribution among fog nodes is a difficult challenge due to the increasing number of IoT devices. In this research, we suggested a new approach for fair load distribution in fog environment. The Q-learning algorithm-based load balancing method is executed as the proposed approach in the fog layer. The objective of this method is to simultaneously improve the load balancing and delay. In this technique, the fog node uses reinforcement learning to choose whether to handle a task it receives via IoT devices directly, or whether to send it to a nearby fog node or the cloud. The simulation findings demonstrate that our approach results a suitable technique for fair load distribution among fog nodes, which improves the delay, run time, network utilization, and standard deviation of load on nodes than other compared techniques. In this way, in the case where the number of fog nodes is considered to be 4, the delay in the proposed method is reduced by around 8.44% in comparison to the load balancing and optimization strategy (LBOS) method, 26.65% in comparison to the secure authentication and load balancing (SALB) method, 29.15% in comparison to the proportional method, 7.75% in comparison to the fog cluster-based load-balancing (FCBLB) method, and 36.22% in comparison to the random method. In the case where the number of fog nodes is considered to be 10, the delay in the proposed method is reduced by around 13.80% in comparison to the LBOS method, 29.84% in comparison to the SALB method, 32.23% in comparison to the proportional method, 13.34% in comparison to the FCBLB method, and 39.1% in comparison to the Random method.