Fog Computing with Distributed Database

Abstract

In recent years, with the progress of IoT, the entry data from various sensors is accumulated on the cloud server and used for various analyses as big data. On the other hand, in order to transfer a large amount of data to the cloud server, there were the problems such as the restriction of network bandwidth, and the delay of feedback control of the sensors. For these problems, Fog computing has been proposed in which the primary processing of the sensor data is performed at the fog node installed near the sensors, and only its processing results are transferred to the cloud server. However, in this method, in the case where the original data of the sensor is required for various analyses at the cloud server, such a data must be transferred additionally. That is, a mechanism is necessary to manage the data of the entire system and to mutually utilize it. In this paper, I propose a data model which consists of three levels: the first level saves the original sensor data and is placed in the fog node; the second level saves the extraction data extracted by the primary processing; the third level saves the analysis results data. The second and third levels are placed in the cloud server. And, by constructing this data model with a distributed database, it can be performed efficiently to refer the arbitrary original sensor data in the fog nodes from the cloud server. Moreover, I implement this reference processing in two ways using MongoDB, which is a kind of NoSQL database, to evaluate this data model. And, I show it is necessary to select the reference way according to the system environment: the network bandwidth, the database performance of the fog node and cloud server, and the number of the fog nodes.