Cost is the constraint most often ignored in interviews, so raising it scores well.
Levers, roughly in order of impact:
- Tiered storage - hot data in memory or SSD, warm on standard object storage, cold in archival storage which is roughly 10x cheaper. Most access is to recent data.
- Retention policy - deleting or aggregating data older than N days often beats any compression scheme.
- Cache to cut egress - a CDN hit avoids both origin compute and origin bandwidth, usually the largest line item for media.
- Right-size replication - a replication factor of 3 for critical data, but 2 or erasure coding for data that can be regenerated.
- Autoscale instead of provisioning for peak - if peak is 3x average, static provisioning wastes about two thirds of capacity.
- Spot or preemptible instances for stateless batch work.
Then state the trade-off you accepted: "Archived data takes minutes to retrieve, which is acceptable for a compliance export but not for a user facing search."