Dynamic Container Placement at Serverless Edge: An Online Convex Optimization Method

Serverless edge computing has garnered significant attention, where container placement emerges as a critical problem owing to dynamic network environment and scarce resources. In this paper, we propose an online convex optimization method to cope with the container placement problem, circumventing sophisticated system modeling and excessive exploration. Within an online learning framework, we develop a strongly adaptive container placement algorithm with regret guarantee to optimize average request delay, where sequential placements are iteratively refined based on historical feedbacks to capture system dynamics. Experimental results illustrate that our proposed method remarkably approaches the performance limit and achieves promising performance gain as compared to both the static optimal comparator and the online gradient descent method.