The mobile edge computing architecture successfully solves the problem of high latency in cloud computing. However, current research focuses on computation offloading and lacks research on service caching issues. To solve the service caching problem, especially for scenarios with high mobility in the Sensor Networks environment, we study the mobility-aware service caching mechanism. Our goal is to maximize the number of users who are served by the local edge-cloud, and we need to make predictions about the user’s target location to avoid invalid service requests. First, we propose an idealized geometric model to predict the target area of a user’s movement. Since it is difficult to obtain all the data needed by the model in practical applications, we use frequent patterns to mine local moving track information. Then, by using the results of the trajectory data mining and the proposed geometric model, we make predictions about the user’s target location. Based on the prediction result and existing service cache, the service request is forwarded to the appropriate base station through the service allocation algorithm. Finally, to be able to train and predict the most popular services online, we propose a service cache selection algorithm based on back-propagation (BP) neural network. The simulation experiments show that our service cache algorithm reduces the service response time by about 13.21% on average compared to other algorithms, and increases the local service proportion by about 15.19% on average compared to the algorithm without mobility prediction.

No abstract

Mobility prediction is a powerful tool for network operators to optimize network performance. From cell level, if network operators know the cells to which the users will be connected in advance, wireless resources can be pre-allocated to improve network performance and better user experience can be provided in location-based services. Many next-cell prediction models and methods have been suggested and implemented. This paper is devoted to next-cell prediction (cell level mobility prediction) in cellular networks, and provides a thorough survey of the prediction schemes and applications. Particularly, a two-level classification methodology was proposed and applied. We first divided the prediction schemes into three categories based on the mobility data used for prediction, i.e. Current Movement State based Approaches (CMSA), Historical Movement Pattern based Approaches (HMPA), and HybriD Approaches (HDA). Prediction schemes in each category were further classified based on the used prediction methods. The typical application scenarios were introduced as well, including handover management, resource allocation, etc. Finally, current challenges and potential trends in the near future were further discussed.

No abstract