Use of multilevel resource clustering for service placement in fog computing environments

The fog is part of the infrastructure that makes up the Internet and can be considered an extension of the cloud. It is a layer between the cloud and the edge, and its topology comprises mostly heterogeneous and resource-limited computing nodes. Because of its proximity to edge devices, the fog is often mentioned as a solution to deploy services with stringent latency requirements. Moreover, the characteristics of the fog are appropriate for the deployment of services with lower computational demands, meaning that lean service models, such as microservices, are an appropriate match. Regarding the placement of services in the fog, an effective technique refers to the use of resource clustering, which organizes resources according to their properties. A common approach has been the use of the geographical location of resources to build clusters that favor the placement of services closer to clients. In this paper, we refine this approach by proposing multilevel clustering, adding to geolocation-based clustering the use of feature-based resource clustering. Thus, besides increasing proximity, we can also improve the matching between service requirements and the characteristics of resources, further improving the performance of applications. We evaluate the approach using iFogSim 2, and the results show performance improvements in excess of 20% in both application flow time and service placement time when compared to pure geolocation-based clustering.