Dynamic Computation Offloading Leveraging Horizontal Task Offloading and Service Migration in Edge Networks

Amidst the progressive proliferation of user petitions in the realm of Internet of Things (IoT), the concept of Edge Computing (EC) is being hailed as a beacon of promise, epitomizing multifaceted dexterity and sound service delivery methodologies. Given the finite resource allocation of IoT devices, it is plausible to assert that those IoT devices that are burdened with arduous workloads might fail to adequately address latency-sensitive requests from users in a prompt manner. In order to surmount the obstacle at hand, we posit a dynamic methodology of computation offloading, whereby a multiplicity of elements are taken into account. These elements include, but are not limited to: (I) The proposition of the intermediate node concept serves to curtail the latency experienced by user requests, via the dynamic consolidation of task offloading and service migration tactics. (II) The intricacies of the intermediate node selection issue are expounded as a multi-faceted Markov Decision Process (MDP) realm, the boundaries of which are contingent on the burden placed upon the incumbent network. As a means of diminishing the expansiveness of the aforementioned MDP sphere and facilitating prompt judgments, a profound reinforcement learning methodology has been custom-tailored. According to the experimental results, this particular modus operandi exemplifies a more pronouncedly impactful curtailment of turn-around time requisite to abide by user requests.