Semi-Optimal MicroService Placement in the Fog with Quality of Service (QoS) Awareness

The proliferation of connected devices has catalyzed a shift towards intelligent systems reliant on real-time analytics for decision-making. While Cloud computing offers a mature service infrastructure, the transfer of vast data volumes for processing introduces challenges such as network congestion and latency, hampering real-time requirements. Edge/Fog computing emerges as a solution, enabling data processing and hosting lightweight learning models closer to users at the network edge. However, unlike the Cloud, which exhibits high resource elasticity with a limited number of resource nodes, Edge/Fog resources are widely distributed with high density, characterized by their heterogeneity and limited capacities. To optimize the deployment of applications with diverse resource and latency requirements, a strategic approach to application placement across Edge/Fog and Cloud computing environments is essential. This strategy must carefully balance application-specific needs, latency sensitivity, and operational cost minimization. To this end, we begin by formulating the placement problem as a linear program, focusing on microservice applications. We then propose a hybrid approach that combines k-means clustering for grouping Edge/Fog nodes (based on physical location) with the formulated linear model to enhance computational efficiency. The linear model is developed using the CPLEX solver and validated using real-world datasets of Edge/Fog node and user distributions. The findings unequivocally demonstrate that the clustering step leads to significantly decreased decision placement times, all while maintaining a high standard of solution quality. These results provide a promising avenue for deploying optimal models directly at the network’s edge, enabling optimal application placement decisions while adhering to problem constraints.