Combine Data Collecting and Task Computing: Progressive Computing in Internet of Things

Abstract

Massive, heterogeneous sensors have been deployed to monitor many objects from multiple aspects, which brings the explosion of sensor data; thus, effective data collection becomes an important issue. Moreover, in emerging advanced IoT applications such as the smart city, computation tasks are released by smartphone users. Such tasks are random and unpredictable, which makes it hard to know in advance what sensor data are needed. Collecting all of the data that might be involved can lead to a huge waste of time, energy, and channel resources. In this article, we study effective sensor data collection problem and the goal is to fetch enough data to get the accurate results of random tasks with minimum cost. We propose a novel progressive computing method, where the sensor data are collected in phases. Specifically, in each collection phase, the server analyzes the data obtained in preview phases and determines what data to fetch in the next phase. Then, the server sends control messages to corresponding sensors by using the downlink communication of Lora. In our computing method, data collection is interwoven with task computing, which enables targeted approaches to the data exactly needed. Two example progressive algorithms are presented, and their performances are analyzed theoretically. Simulation results show that our algorithms can reduce the cost remarkably without loss in the accuracy of the computation results.