Datacenter Exit Program Playbook

Datacenter Exit Program Playbook

A datacenter exit is a fixed-deadline program to move every workload out of one or more owned facilities and shut down the physical estate. The driver is usually a lease expiry, a hardware refresh you no longer want to fund, or a board-level cloud mandate. Unlike an open-ended migration, an exit has a hard date: if the building is not empty, you keep paying for it.

Phase-by-Phase

Discovery and Inventory. You cannot migrate what you cannot see. Build a complete inventory of servers, storage, network appliances, and software licenses, then map application-to-application dependencies. Tag each workload with a disposition from the 7 Rs (rehost, replatform, repurchase, refactor, retire, retain, relocate). Retiring dead workloads is the cheapest win available.

Landing Zone and Network. Stand up a cloud landing zone with account structure, identity, logging, and network guardrails before any workload moves. Establish hybrid connectivity (Direct Connect, ExpressRoute, or Interconnect) so the old and new estates can coexist during migration.

Wave Migration. Group workloads into waves by dependency cluster, not by team convenience. Migrate, validate, and cut over one wave at a time. Keep a tested rollback for every cutover until the source is decommissioned. Rehost simple VMs first to build momentum; replatform or refactor the workloads that justify the effort.

Decommission and Verify. The exit is only real when the hardware is powered off, wiped, and removed, and the facility contract is terminated. Produce an exit certification that proves zero residual footprint.

Team and Roles

A program lead owns the deadline. Architects own dispositions and the landing zone. DevOps and SRE run the migration factory. Security signs off on each wave. DBAs handle data migration, the highest-risk activity. Product owners accept cutovers for their applications.

Risks and Mitigations

Hidden dependencies are the classic killer; mitigate with dependency discovery tooling and traffic analysis, not interviews alone. Data migration loss is mitigated with checksums and parallel-run validation. Deadline slippage is mitigated by retiring aggressively and by tracking burndown weekly. Dual-running cost is mitigated by sequencing decommission immediately after each wave proves stable.

Success Criteria

The datacenter is fully vacated by the deadline, no residual assets remain, run-rate cost drops versus the on-prem baseline, and 100% of in-scope workloads are migrated or retired.

Tooling

Use cloud-native migration services for discovery and replication, Terraform for the landing zone, Kubernetes for replatformed workloads, Datadog for cross-estate observability during the transition, and Vault for secrets that move with the workloads.