Shepard: Dynamic Placement of Microservices in the Edge-Cloud Continuum

We present Shepard, an innovative microservice placement approach tailored for edge-assisted cloud infrastructures. Shepard dynamically migrates application services between the edge and cloud to harness optimal performance gains. This approach is structured around three core components: (1) a resource manager for monitoring available edge resources and the evolving demands of applications, (2) an optimization module that transposes the service placement dilemma into a labeled-graph cut challenge, aiming to identify the most advantageous cut given a set of parameters, and (3) a deployment module tasked with adjusting service placement in response to shifts in the optimal graph cut's position. Our implementation of Shepard underwent rigorous testing in two distinct case studies. In the inaugural study, Shepard managed energy for a solar-driven edge within an agricultural IoT framework, resulting in a striking 79% elevation in service reliability and availability compared to a conventional static service placement strategy. For our subsequent study focusing on cost-effectiveness within a ride-hailing application, Shepard facilitated a substantial 45% slash in application deployment expenses, all the while maintaining comparable performance levels to a standard dynamic service placement technique.