Interconnection Solutions for Distributed Data Centers: Storage, Layer‑2, and Layer‑3 Networks

Distributed data center deployment enables dynamic allocation of computing resources across data centers using technologies such as VMware vMotion for virtual machine (VM) migration and high‑availability server clusters for application‑level disaster recovery. An application can serve users from any data center while keeping its IP address unchanged, but it appears in only one data center at a time.

1. VM Cross‑Center Dynamic Migration The most common scenario relies on VM memory and CPU state replication between physical hosts, requiring shared storage (NAS or SAN) or active‑active storage to provide the same VM image to both source and destination hosts.

Challenge 1: Storage Latency Sensitivity Long distances increase storage I/O latency. Active‑active storage solutions (e.g., EMC VELEX Metro) create a logical storage device accessible locally at each data center, reducing latency compared to shared storage. FC SAN, iSCSI SAN, and NAS interconnects are typically built on TCP/IP.

Challenge 2: Maintaining Runtime State After Migration To keep the VM’s IP address and TCP sessions intact, both data centers must share the same Layer‑2 network. Technologies such as MAC‑over‑IP (e.g., H3C EVI), VPLS, and DWDM/Dark Fiber can extend VLANs across sites, with EVI offering gateway separation, MAC learning, and ARP proxy features.

Challenge 3: Optimizing Layer‑3 Path Selection After migration, the optimal ingress and egress routes may become sub‑optimal if routing announcements do not change. Solutions include dynamic DNS/GSLB, Route Health Injection (RHI) on load balancers, and gateway‑separation with synchronized VRRP configurations to ensure traffic prefers the nearest data center.

2. Server High‑Availability Cluster Across Centers HA clusters present multiple physical servers as a single logical server with a floating IP. Cluster software from vendors (HP, IBM, Microsoft, Veritas) requires shared or active‑active storage and a common VLAN for heartbeat links. The same Layer‑3 path optimization issues apply as with VM migration.

Conclusion The three interconnection models—storage network, Layer‑2 network, and Layer‑3 network—are essential for enabling VM migration and HA clustering in distributed data centers. Selecting the appropriate technology depends on the specific business requirements, physical distances, and desired performance characteristics.