Glossary
Data fabric
A data fabric is an architecture that connects data spread across storage systems, applications, clouds and sites so it can be found, governed and moved under one set of rules. It adds a shared metadata and policy layer over existing storage rather than gathering everything into one repository.
The problem it addresses
Large organizations hold data in several clouds, a few data centers and dozens of applications, each with its own access rules and naming. The data is rarely missing. What is missing is a reliable answer to what exists, who owns it, where it sits and whether a given team may use it. A fabric treats that as a coordination problem and leaves the bytes where they are.
Layers of a fabric
- Existing storage. Arrays, object stores, file systems and cloud buckets, which the fabric does not replace.
- Metadata catalog. What exists, where, in what format, owned by whom and tagged how. Every other layer depends on it.
- Policy and identity. Access, retention and residency rules applied identically wherever the data sits.
- Interfaces. How applications and pipelines discover and request data without knowing which system holds it.
Automation spans the layers and applies policy as conditions change, for example restricting access to a dataset when its classification is raised.
Fabric, mesh and consolidation compared
| Approach | What it is | Where data sits |
|---|---|---|
| Data fabric | A technology layer of catalog, policy and interfaces | Where it already is |
| Data mesh | An organizational model in which domain teams own data as products | With each domain |
| Consolidation into a data lake | One shared repository | Copied into the central store |
Consolidation suits some datasets and fails for others, because large volumes are costly to move, the effect called data gravity, and because regulation often keeps data in the country or system where it began. A fabric accepts that distribution and makes it manageable. Moving data, under data mobility, happens when policy calls for it, not by default.
A fabric also differs from a lake in what it costs to adopt. A lake requires data to be copied in before it is useful. A fabric requires the catalog and policy layers to reach every source system, which shifts the effort from moving bytes to integrating metadata.
Why does metadata decide the outcome?
A fabric can only report what its sources record. Laid over systems with inconsistent tags and no named owners, it produces a faster view of the same confusion. Ownership, naming and classification have to exist at the source first, which is the same discipline AI data management relies on when training data must be traced.
Frequently asked questions
Is a data fabric a product or an architecture?
An architecture. Tools implement parts of it, most often the catalog and policy layers, but no single product supplies every layer over arbitrary storage.
Does a data fabric move data?
Only when a policy requires it. The fabric tracks and governs data in place, and copies, moves or tiers it according to rules.