Cutover Planning for an 80 TB Heterogeneous Migration
A logistics company is migrating 80 TB of business data from an ageing EMC VNX to a new OceanStor Dorado. The estate includes 25 SQL Server instances and 300 VMware VMs; the main cutover window is 10 hours on a Saturday. The integrator proposes a host-based copy with robocopy, a 6-hour rehearsal, and a single cutover. In the rehearsal, the initial copy took 14 hours and the incremental copy took 3 hours. The go-live is three weeks away. The business can tolerate four hours of downtime for non-critical systems and one hour for order processing, but not a data loss beyond 15 minutes. The project manager must choose an array-based migration, host-based migration, or a hybrid approach, and define rollback.
1. Evaluate the integrator's plan and recommend a migration approach. Include how to handle the 14-hour initial copy.
Read the worked answer
The robocopy plan is not credible. A 14-hour initial copy cannot fit a 10-hour window, and a 3-hour incremental leaves too little time for validation and rollback. Treat the initial copy as a pre-stage, not cutover work.
Recommended approach: hybrid. Run a host-based or migration-appliance copy of the bulk 80 TB two to three weeks before cutover, with throttling and restart capability, then run repeated incremental synchronisations. In the 10-hour window, stop applications, run one final incremental delta, validate, switch and start applications. If the VNX is on Huawei's supported virtualization matrix, array-based takeover or SmartVirtualization could reduce host work, but do not assume heterogeneous array-based migration is supported; confirm the support matrix first. Split into waves: order processing first with its one-hour tolerance, then non-critical systems with four hours. Keep the source unchanged and bootable until business acceptance.
2. Build a cutover plan with milestones and rollback triggers for the 10-hour window.
Read the worked answer
Cutover plan: T-14 days pre-copy and baseline; T-7 repeat incremental and test restore; T-3 freeze non-essential changes; T-1 final backup and go/no-go. In the window: 00:00 stop order applications; 00:15 final incremental; 02:00 validate row counts, file checksums and application transactions; 03:00 switch and start; 04:00 business smoke test. Set checkpoints at 2, 4, 6 and 8 hours. Rollback triggers: final incremental not complete by hour 3; data validation mismatch; critical application not transactionally consistent by hour 6; or order-processing downtime exceeding one hour. Rollback means restart source applications, revert DNS and records, and keep the target for analysis. Communicate status every hour.
3. List the top three project risks and the acceptance criteria you would present to the business.
Read the worked answer
Top risks: (1) data inconsistency between SQL and file shares, mitigated with application-consistent snapshots and reconciliation; (2) cutover overrun, mitigated with pre-staging, rehearsals and go/no-go checkpoints; (3) performance regression on the new array, mitigated with a pre-go-live benchmark and tuning. Acceptance criteria: all 25 SQL instances pass DBCC and application transaction tests; 300 VMs boot and pass smoke tests; file checksums match; order-processing downtime is at most one hour and non-critical downtime at most four hours; no data loss beyond the last committed transaction, with the 15-minute backup RPO verified; performance meets baseline; the business owner signs off. Keep a five-day hypercare period.