An agent is a computer system that is situated in some environment and can act autonomously. Fog computing is a computing paradigm in which computation is executed on a computing node that resides near or within an edge network, rather than a cloud. As the demands of Internet of Things (IoT) applications increase, increasing attention is paid to agent systems and fog computing. Since typical IoT devices reside in dynamic and challenging environments, the autonomous and suitable feature of being an agent is attractive. Also, fog computing is necessary to defeat the network delay in reactive IoT applications. Thus, the title of the paper must be interesting to IoT researchers.

This paper proposes an architecture consisting of three layers: a physical layer where physical IoT devices exist, an intermediate layer where computing nodes exist, and a cloud layer. Each computing node is composed of virtual objects representing physical IoT devices, a gateway, and an agent-server. The cloud layer includes virtual objects of computing nodes.

The proposed architecture is described too briefly to understand its details. It appears that the data analysis is done in the cloud layer, but it is not clear how the cloud-based data analysis and swarm intelligence are related. In addition, it is not clear how each agent interacts with its peers to achieve swarm intelligence in the fog computing environment.

This paper presents three IoT applications using their architecture, noise pollution mapping, urban drainage networks, and smart street. The motivation of the first application is that the proposed architecture can account for the real-time variation of the noise level instead of estimating long-term average noise levels. However, why is it important to measure real-time noise variation? It seems to be reasonable to make plans against noise pollution based on long-term average noise instead of the real-time variation. This paper does not explain the motivation sufficiently. In the second application, the scheme should be compared with a centralized drainage control system because the main idea of the proposed scheme is using distributed agents and fog computing; it is not clear if the suggested system is effective.

The technical depth of the paper is disappointing. Nevertheless, the approach of combining agents and fog computing for IoT application is interesting.