Before a scale-up, a migration or a multi-year commitment, a platform’s resilience is validated through engineering, not on paper. I assess the scalability and reliability of your data and AI platforms, and I give a reachability verdict on your RTO and RPO: measured restore throughput, replication lag under partition, dependency ordering during recovery. Never a documentation or compliance exercise.
Scalability and reliability
Growth scenarios confronted with real load, stress tests pushed to the limits, the blast radius of every failure mode, high-availability designs verified by measurement rather than by the architecture diagram. Reliability reads in operations metrics: what alerts, what degrades silently, what never surfaces at all.
Disaster recovery: RTO and RPO
A backup restores data; disaster recovery restores a service. I measure restore throughput on your real volumes, verify backup integrity through actual restores, and time the recovery with its dependency ordering: databases before applications, network before data, keys before everything. The verdict is quantified: this RTO is reachable, that one is not, and here is what would have to change.
Failover testing
A recovery plan that was never executed remains an assumption. I run failover tests under realistic conditions: a site down, a datacenter lost, a full restore from backups. Every test produces the list of gaps between the plan and reality, with fixes ordered by cost of correction. The topic is covered in an article on MinIO site replication and a multi-site S3 upgrade on Ceph.
This is neither documentation audit nor regulatory compliance: it is engineering validation, backed by measurements.
FAQ
How do you verify that an RTO and an RPO are actually reachable?
By measuring, not by reading: restore throughput on real volumes, a failover executed with its dependency ordering, replication lag observed under partition. An RTO claimed without a documented failover test is an assumption, not a guarantee.
What is the difference between a backup and disaster recovery?
A backup restores data; disaster recovery restores a service, with its dependencies, its clients and its writes to replay. An intact backup with an unreachable RTO is still a prolonged outage: DR is judged on the complete chain, not on the backup medium.
How often should failover be tested?
After every major change to architecture, replication or backups, and periodically otherwise, because platforms drift even when the plan does not. One test a year is the floor; critical platforms test at every major release.