Real-Time IOT Applications Offloading Decision Strategy

With the advancement of Internet of Things (IOT) around the world, different devices connected to internet generate enormous data from heterogenous resources with different processing capacities. Management of these devices and processing data of this kind is a challenge for the service providers. Scheduling all these requests from these devices to virtual resources in cloud paradigm might be complex. The main challenge for the service providers is to assign appropriate resources to heterogeneous requests comes from different devices with different processing capacities because there are delay sensitive tasks (e.g., Healthcare applications, Smart cars, Autonomous vehicles) which need to be processed and executed on demand at the edge locations of the service providers. Computational offloading strategy is a promising solution to solve these kind of problems by computing them on edge servers by choosing appropriate computational sources and resource constraints rather than sending them to virtual resources in the cloud environment. Many authors proposed different task offloading mechanisms by posing resource constraints and used various metaheuristic approaches yet this problem persists due to dynamic nature of nodes, energy constraints and in turn poses challenges for the service providers. Therefore, to minimize consumption of energy and latency incurred due to the inappropriate assignment of upcoming IOT tasks, in this research we proposed a new offloading mechanism in two stages. In the first stage, a task classification mechanism is induced which classify all upcoming tasks with respect to delay sensitive based on deadline constraints while tracking the availability of computing nodes at the edge locations to make a decision either to schedule tasks to virtual resources in cloud or to the nodes at the edge locations. In the second stage, a task scheduling mechanism modeled by a reinforcement algorithm named Asynchronous Advantage Actor Critic (A3C) is used to optimize generated schedules. Extensive simulations are conducted using SimPy. Proposed real-time IOT offloading mechanism using A3C (RIOTOA3C) compared against existing DQN, A2C algorithms. Results of the proposed approach shown significant improvement in minimizing energy consumption, latency.