A reinforcement learning algorithm for rescheduling preempted tasks in fog nodes

The fog server in a fog computing paradigm extends cloud services to latency-sensitive tasks by employing fog nodes (FNs) near user devices. The resource-constrained FNs face the challenge of meeting stringent deadlines of latency-sensitive tasks. The completion deadline of such tasks becomes critical on preemption. Task preemption is unavoidable in uncertain events, such as FN hostility, overloading, and mobility of the host FN or the user device. Rescheduling the task that is likely to face preemption is a better solution than terminating it. This paper proposes a rescheduling algorithm for the fog server to reschedule preempted tasks to FNs that can serve them to completion within their expected time. The rescheduling algorithm aims to attain a rescheduling list that guarantees the task deadline requirements. The brain-inspired rescheduling decision-making (BIRD) algorithm proposed in this paper uses the actor-critic reinforcement learning method for rescheduling preempted tasks to FNs. It mimics the decision-making model of the human brain to control voluntary motor activity. It guarantees the deadline requirement of the preempted task by ensuring the optimal performance of the FN through load balancing while rescheduling the preempted tasks to FNs. Experimental evaluation shows that the BIRD algorithm offers better FN selection than other scheduling policies such as first come first served (FCFS), greedy task allocation, task allocation based on least laxity, shortest job first (SJF), and earliest deadline first (EDF).