You are planning a network design update for a shared infrastructure environment that carries AI training jobs, storage replication, and east-west service mesh traffic on the same fabric. The goal is to improve isolation and predictability, but the environment is highly performance-sensitive and any segmentation choice could introduce congestion, operational complexity, or unexpected bottlenecks.
How would you segment traffic for mixed workloads such as AI training, storage replication, and service mesh without compromising throughput?