Machine Learning Methods in Tasks Load Balancing Between IoT Devices and the Cloud

Nowadays, with the ongoing, wide scale digitization and development of AI in pursuit of automation, the IoT industry becomes one of the very important parts in this process. The development of IoT devices computational capabilities, as well as the massive amounts of data that is being generated by them, create a need for methods to load balance workloads efficiently. Since the IoT devices are receiving more processing power, it becomes important to leverage that power for executing curtain tasks inside an IoT ecosystem itself, rather than delegating to the Cloud. The paper focuses on exploration of the existing solutions and offers an alternative concept. The goals of the research are: 1) to understand what mechanism would allow to distribute tasks among IoT devices and Cloud servers, and what are the potential criteria for that, 2) to see, how feasible it is to distribute tasks between devices, matching them by task runtime complexity and device performance. The solution we propose is to distribute tasks based on several parameters, such as runtime complexity, energy and memory required to process a task, and historical data from executed similar tasks. The parameter estimation involves device performance testing, executed task data aggregation, and application of machine learning (ML) to approximate task runtime complexity. The results from conducted experiments show that the proposed concept is a viable solution and provide opportunity for further research.