Build a requirement-to-test matrix
Link every requirement to a procedure, expected result, owner and evidence. “The server is installed” does not prove performance or recovery.
Verify the baseline
- Actual models, serial numbers, licenses, versions, cabling, naming, time synchronization and access.
- Redundant power, network paths, controllers and disks.
- Supported firmware and software combinations.
Test the workload
Use realistic data volume, concurrency, request size and duration. Record peaks, latency, queues, errors and resource use, not only averages.
Trigger controlled failures
- Lose a network path, power source or cluster node.
- Fail a disk or controller within the agreed design.
- Remove an external dependency such as DNS, identity or connectivity.
- Restore from backup and measure actual RTO and RPO.
Verify observability and operations
The event should create a meaningful monitor and log entry, reach the right on-call path and leave enough evidence for diagnosis. Accept as-built diagrams, configuration backups, operating procedures, support contacts, SLA boundaries and known limitations.
Close with an evidence-based protocol
Record date, version, inputs, result and evidence for each test. Give every deviation an impact, owner, due date and explicit disposition.