Cloudlet Overload Prediction Based on Deep Learning

Abstract

Cloudlet is a typical model of edge computing, which has received widespread attention as it can better meet users' latency-sensitive business needs compared with the traditional cloud computing models. Accurate prediction of cloudlet overload is an important prerequisite for designing a more effective cloudlet task migration strategy. In this paper, we first build a multi-cloudlet load model. By designing the basis for judgment of cloudlet overload and task migration strategy, a dataset that can be used to predict the load of cloudlet using supervised learning strategy is generated. Then a deep learning-based cloudlet load prediction model is designed and trained to predict whether the cloudlet will be overloaded by describing a series of parameters such as the arrival of user requests, the required network bandwidth, and computing resources. The experimental results validate the effectiveness of the model.