Running thousands of snapshotted VMs pushes Linux networking to its limits. This talk covers battle-tested strategies balancing performance, isolation, and efficiency across large-scale topologies. We explore Geneve tunneling for scalable overlay networks, eBPF-accelerated networking for line-rate performance, and stateless NAT without connection tracking overhead. Prefix-based iptables/nftables rules scale to thousands of instances, while namespace pooling minimizes snapshot operation costs. We also address RTNL lock contention and its mitigation. Attendees will learn to architect network stacks for snapshotted workloads where VMs are cloned, restored, and migrated while maintaining security and predictable performance.