A Context-aware Approach to Task Scheduling for Time Series Data Prediction in Mobile Edge Computing

Context-aware time-series data prediction is crucial and ubiquitous in cloud-edge collaborative storage in Mobile Edge Computing. Most existing studies aim to reduce energy consumption or improve performance in different aspects by computation offloading among the MEC nodes. In the MEC scenario, the mobile edge devices generate and send time series data at a high frequency. The edge nodes will process the collected data to support real-time tasks and periodically migrate the compressed local data to Cloud Data Center(CDC). However, when it comes to reality, it is crucial to locate the data and decide the task execution strategy. In this paper, we first design a management system to manage the metadata and the data collected from mobile edge devices. Based on the management system, we propose a context-aware approach to task scheduling for time-series data prediction in MEC, using the task contexts to locate the data sources, generate the task execution strategy, and choose the targets of result forwarding. For the data prediction, we adopted the LSTM model to predict future time-series data considering the performance. We evaluate the feasibility and effectiveness of the proposed approach on an Automatic Identification System (AIS) dataset. The results illustrate that the proposed approach can effectively schedule the context-aware time-series data prediction tasks in MEC.