Reliability has a price. So does overengineering.

Situation

Infrastructure reliability carries a cost, but not every workload creates the same business impact when it fails. Treating every component as equally critical was spending without enough attention to what customers actually needed.

Context

Some stored assets supported a secondary feature, HyperZoom. A temporary outage would be inconvenient, but the feature could be disabled and the assets rebuilt over time. That failure mode was materially different from losing a core service.

Decision and approach

Those workloads were moved to a less redundant storage provider. The choice was not based on price alone: the likely failure, customer impact, recovery path and rebuild time were considered together. The system allowed graceful degradation by temporarily switching off the affected feature while protecting more important behaviour. Reliability investment could then be concentrated where an interruption would carry greater consequences.

Outcome

The architecture matched infrastructure cost to business value. It accepted a bounded, recoverable risk instead of paying for maximum redundancy everywhere, without pretending the trade-off had disappeared.

Lessons and principles

Cost control is an engineering and commercial judgement, not a blanket demand to spend less. Define the impact the business can tolerate, preserve a recovery path, and buy the reliability each workload genuinely justifies.