Online Learning and Resource Allocation for User Experience Improvement in Mobile Edge Clouds

Abstract

As the cloud is pushed to the edge of the network to promote the development of mobile applications, resource allocation for user experience improvement in mobile edge clouds is facing multiple challenges. In order to serve mobile users, resource allocation in mobile edge clouds needs to be adjusted continuously. Considering that the available resources in mobile edge clouds are generally limited, we propose to optimize resource allocation by minimizing the time-average complaining probability. Specifically, we advocate a feedback-based online resource allocation system, which can effectively improve the overall user experience in mobile edge clouds through online resource allocation. The system runs in a closed-loop fashion: we establish a classifier model to predict user experience, use the prediction results as the target of resource allocation model to guide the resource allocation explicitly, and then adjust the classifier model based on the updated overall dataset. The online resource allocation problem turns out to be a stochastic problem, for which we design an online queue resource allocation algorithm (OQRAA) supported by Lyapunov optimization technique. The proposed system enables us to explore the quantitative relationship between user experience and feature values and leverage this relationship to find the optimal resource allocation policy immediately upon users’ arrival. The numerical results show that the proposed system has an improvement of 200% in reducing the time-average complaining probability compared with the baseline algorithm.