μLayer: Layer-Aware Microservice Placement to Maintain QoS in Edge Computing Under Bandwidth Uncertainty

Abstract

Container-based microservices provide a lightweight and scalable foundation for edge computing, where layer sharing improves storage and throughput efficiency. However, the unpredictable nature of edge networks leads to variable layer download times and thus bandwidth uncertainty. This article addresses the problem of microservice placement under bandwidth uncertainty by incorporating a quality of service (QoS) constraint that ensures timely start-up with high probability. The resulting model is a complex nonlinear stochastic optimization problem. By employing a series of mathematical lemmas, we reformulate it as an integer linear program (ILP), which can be solved with conventional computational tools. Furthermore, we propose <inline-formula> <tex-math notation="LaTeX">$\mu $ </tex-math></inline-formula>Layer, which relaxes the ILP to a linear program (LP) and applies iterative rounding to obtain high-quality solutions with improved efficiency and scalability. Experimental results demonstrate that the <inline-formula> <tex-math notation="LaTeX">$\mu $ </tex-math></inline-formula>Layer substantially reduces execution time while maintaining placement quality close to the optimal solution. In addition, compared with the baseline methods considered in this study, <inline-formula> <tex-math notation="LaTeX">$\mu $ </tex-math></inline-formula>Layer improves the total placement value of microservices by up to 25% and increases the probability of meeting the expected layer download time by 15%, which demonstrates its ability to handle bandwidth uncertainty more effectively.