Out of the hurricane zone: a 1,000‑person Miami distributor moves from two server rooms to Google Cloud

In South Florida, concentration is the risk. This importer and distributor ran everything from a main server room and a backup room less than ten miles apart, in the same evacuation zone. A tripped transfer switch and a generator that ran out of fuel had already cost it a day, and the backup site had never passed a full failover test. Its ERP, warehouse system and customs application had to be available from early morning until midnight, six days a week, across three time zones. Meanwhile the mail servers were reaching end of support, a virtualization renewal quote had multiplied, month-end close took eight business days because reports were rebuilt by hand from four systems, and a single VPN appliance served a thousand people. The board asked for four things: no single building, region or grid could take the business down, tested recovery of the core systems, an exit from both server rooms, and access from anywhere without a VPN.

We began with an inventory and a framework, not a product list. A 30‑day discovery mapped every server and dependency, and each workload then got one of five dispositions: retire, repurchase as software as a service, rehost, replatform or refactor. Each choice was scored on recovery goals, operating effort, three-year cost, migration risk, existing skills and fit with the data roadmap, and written into a short decision record that the client's IT director signed. When the project ended, the client's staff could extend those decisions rather than reopen them.

The target is two Google Cloud regions outside Florida's weather and power grid, joined to the offices and the port warehouse by two private links through different carriers, with encrypted VPN as backup. Mail and files moved to Google Workspace, and access control based on identity and device health replaced the VPN. Databases moved to managed services with cross-region replicas, and the order portal was rebuilt on containers behind a global load balancer with a web application firewall. A small on-site gateway keeps label printing alive if the link fails.

The sequence mattered. We ordered the network links first, then moved the tightly coupled core, the ERP and the servers it depends on, in the weeks before 1 June with a temporary VMware landing zone, and converted those servers to native VMs afterward. A change program ran alongside: one trained champion per team, training in English and Spanish on every shift, and floor walkers after each wave. All figures in this case are modeled and the engagement is illustrative, and Google Cloud product details should be checked against current documentation before any real plan.

The work ran in seven waves over about fifteen months: roughly 300 virtual machines, 40 databases, 100 TB of files and 1,000 mailboxes. The goal was simple to state: by the start of the Atlantic hurricane season on 1 June, every tier-1 system was in the cloud. Databases, the order portal and the last servers followed, and the server rooms closed in the new year. Recovery is tiered and drilled each quarter. The replica in the second region is designed for about a minute of data loss and five minutes to recover; immutable backups give a day of data loss and a few hours to recover but survive an attacker with administrator rights, which is why the case keeps both.