Industrial Microservice Redeployment Based on Communication Resources Sharing in Kubernetes
Year: 2026
Type: article
Abstract: In industrial edge computing environments based on Kubernetes, the scheduling of microservices often focuses on computing resource allocation (e.g., CPU and memory), while neglecting the coupling between communication resources (image pulling, inter-service traffic, and bandwidth sharing).This oversight leads to prolonged service startup times and severe network contention, particularly in large-scale industrial scenarios.To address this issue, we propose an end-to-end industrial microservice redeployment framework that jointly optimizes communication and computing resources.First, we model the overhead of container image pulling and the communication latency of service chains under bandwidth constraints.Then, we formulate the redeployment problem as a mixed-integer linear programming (MILP) problem aiming to minimize the comprehensive system overhead, including image pulling delay and service chain communication delay.To solve this non-convex problem efficiently, we introduce a successive convex approximation (SCA) framework to linearize the coupling terms, achieving an approximate optimal solution.Experimental results on a 5-node Kubernetes cluster show that our proposed strategy reduces the system comprehensive overhead by 32.5% compared with mainstream scheduling algorithms (DS, GDS, and LS), effectively improves image pulling efficiency, and alleviates network link congestion.
Authors: Hao Feng, Jian Zhao, Mingkai Zhen
Institution: Tongji University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7215123160