Reinforcement Learning based Fog and Cloud Resource Allocation for an IoRT-aware Business Process

Abstract

IoRT-aware BP aims to promote the businessprocess (BP) within robotics and IoT capacities.This incorporation ensures machine-to-machine (M2M)communication and the automatic execution of tasksby using robot devices. Nonetheless, the execution ofthis process inside the enterprise may be costly dueto the consumed resources, the need for computationalcapacity, etc. To close these gaps, the businessprocess outsourcing (BPO) strategy can be carried out tooutsource the IoRT-aware BP to external environments(e.g., Cloud, Fog, etc.). To profit from outsourcing, anenterprise should identify suitable resources to ensureoptimal process execution. The selection of resourcesis known in the literature as resource allocation (RA).The RA problem is described in this work using theMarkov Decision Process (MDP), and it is resolved usingreinforcement learning (RL). The proposed approachrelies, on one hand, on Q-learning as an RL algorithm,and on the other hand, it considers the extensionof the ifogSim tool to support the process executionusing Fog and Cloud resources. The obtained resultsare promising in terms of response time regarding thescale-up of the considered resources. Furthermore,the experimental results show that our approach offersa substantial advantage in optimizing the performanceof RA, which confirms its usefulness and relevancecompared to other common methods.