Database / Milvus Vector database Interview questions
What is the difference between Milvus's tiered storage and traditional single-tier storage?
Traditional single-tier storage treats all of a collection's data uniformly, the same storage medium and access path regardless of whether a given piece of data is queried constantly or almost never. Tiered storage instead classifies data based on actual access patterns and places "hot" (frequently accessed) data in faster, more expensive storage/memory while "cold" (rarely accessed) data moves automatically to cheaper storage tiers.
| Single-tier storage | Tiered storage |
| Uniform cost and performance for all data. | Cost and performance vary by data's actual access frequency. |
| Simpler to reason about, but doesn't optimize for real-world skewed access patterns. | Automatically adapts to hot/cold access patterns without manual intervention. |
| Can significantly overpay for rarely-accessed data held in premium storage. | Reduces overall infrastructure cost by matching storage tier to actual usage. |
This matters because real-world vector datasets are rarely accessed uniformly: a small fraction of entities (recent data, popular items) typically receives the bulk of query traffic, while a long tail of older or less relevant data is queried rarely, if ever. Tiered storage lets a system capture the cost savings of that skew automatically, rather than requiring every application to manually implement its own hot/cold data management strategy.
More Related questions...
