Dynamic Service Caching In 5G-Enabled MECs With Bursty User Demands

With the rapid development of modern network technology,the explosive growth of mobile data traffic has caused increasingly high demand for network latency in mobile applications,so the network service providers face enormous challenges in reducing service delays and bandwidth pressures.5G is envisioned as a key technology to satisfy the huge demand in AI and high-performance services across all industries and businesses with the characteristics of high speed,low latency,and multiple connections.Mobile edge computing(MEC),is a crucial part of the 5G technology and provides extreme low-latency services in the next generation 5G access networks.However,computing resources and storage capacity are limited in 5G-enabled MEC,so how to efficiently use these computing resources to provide services for more mobile users is a challenge facing 5G networks.In 5G-enabled MEC,network service providers deploy computing resources can cache their services from remote data centers to base stations to serve user tasks in their close proximity,thereby reducing the service latency.However,mobile users usually have various dynamic hidden features,such as their locations,user group tags,and mobility patterns.Such hidden features normally lead to uncertainties of the 5G-enabled MEC,such as user demand and processing delay.In this paper,we investigate the problem of dynamic service caching and task offloading in a 5G-enabled MEC with user demand and processing delay uncertainties.We first propose an online learning algorithm for the problem with given user demands by utilizing the technique of Multi-Armed Bandits(MAB),and theoretically analyze the regret bound of the algorithm.We also propose a novel architecture of Generative Adversarial Networks(GAN)to accurately predict the user demands based on small samples of hidden features of mobile users.Based on the proposed GAN model,we then devise an efficient heuristic for the problem with the uncertainties of both user demand and processing delay.We finally evaluate the performance of the proposed algorithms by simulations based on a realistic dataset of user data.Experiment results show that the performance of the proposed algorithms outperform existing algorithms.