Machine Learning Assisted Heuristic Approach for Optimal Task Deployment in Hybrid Cloud / Edge Environments

Dynamically allocating tasks in an IoT environment can benefit from resource state prediction, task requirements and various forms of machine learning. This paper proposes a distributed task allocation scheme, for IoT networks in a hybrid Edge-Cloud platform, which relies on a combination of machine learning and heuristic methods in an auction-based system with an objective to provide near-optimal task allocation. The proposal is tested on several workloads, developed using realtime data from IoT and Cloud computing environments, and the results are compared against various simple-heuristic and complex-optimization-based allocation schemes. The results show that the proposal approaches a near-optimal allocation of tasks to resources with low decision-making overhead owing to the lightweight learning system.