If the Mimir in question is Grafana Mimir, compare long-term Prometheus-compatible metrics storage with the same write and query workload. Start from the Mimir alternatives page and the official architecture guide.
- Send a representative remote-write stream with real label cardinality.
- Run the alerts and PromQL queries operators rely on; compare results and p95 latency.
- Test tenant isolation, retention, backfill, and recovery after an ingester failure.
- Measure object-storage requests, compute, and operational overhead.
A cheaper storage number is not enough if queries become slow or missing data goes unnoticed.