Collaborative learning-based Schema for Predicting Resource Usage and Performance in F2C Paradigm
Souvik Sengupta, Jordi García, Xavi Masip‐Bruin · 2020
Better resource utilization is a continuous demand for smart computing paradigm. Fog-to-Cloud (F2C) is emerging to satisfy that need. Adopting machine-learning (ML) techniques help F2C to priorly forecast system resource usage and performance, for keeping its promise towards better resource management. Performing ML techniques for predicting the resource usage and performance needs historical data of system resources. Processing over that data for extracting the knowledge is one of the essential steps in ML techniques. Based on the obtained knowledge, prediction model is built to perform forecasting operation. This course can be done either in a logically centralized location (cloud) or distributed locations. Due to data privacy issues and bandwidth limitation, it is often an impractical and massively resource-consuming job to continuously send those historical data to the centralized location. Also, the local execution of these operations reduces service latency for any time-sensitive applications. Considering these and observing F2C computing facilities, we realize to build a new architectural framework for adaptively managing F2C resources. In this framework, we implemented a globally coordination-based distributed learning and prediction mechanism for accurately forecasting the resource usage (i.e., RAM, CPU, Disk) and execution-time (performance) for achieving some tasks. Finally, performing comparative tests between our proposed framework and centralized learning and prediction based computing framework, shows our proposed framework outperforms over the centralized framework, in terms of prediction accuracy. Also, we found that in our proposed framework, the overall bandwidth utilization reduces by one-third, and data transmission time becomes lower by half to the centralized framework.