Towards efficient affinity scheduling for complex multi-agent systems in containerised cyber-physical environments

Cyber-physical multi-agent systems require efficient scheduling of heterogeneous, communication-intensive agents. However, Kubernetes often neglects frequent inter-agent communication and affinity dependencies, causing excessive cross-node latency and resource imbalance. This paper formulates affinity-constrained container scheduling, establishes its mathematical model, and proves the problem to be NP-hard. We propose MSAS-CMAS, combining hierarchical clustering, greedy node matching, and worst-fit resource adjustment. Experiments on an 8-node cluster with 120 pods show that MSAS-CMAS reduces communication latency by 42.45% and the number of heavily involved cross-node communication nodes by 62.5% compared with Kubernetes.>