CDMI

CDMI is the storage industry’s own standard for talking to cloud storage over HTTP, and the one the commercial clouds ignored in favor of copying Amazon S3. It is more ambitious than S3: it is not only a way to read and write objects but a way to tell the storage system how to manage them, and a way for the client to discover what the system can do.

• Three object types - Data objects (files), container objects (directories, which can also be addressed by object ID independent of path), and queue objects for ordered delivery.
• Metadata drives management - A client sets retention, placement, replication and lifecycle requirements as metadata, and the system reports back what it actually did.
• A capabilities object - Every container exposes cdmi_capabilities, a resource the client reads to learn which operations, metadata and features the server supports before trying them. Capability discovery is in the protocol.
• Multi-protocol namespace - CDMI is designed to sit alongside NFS, SMB and object protocols over the same namespace, and to carry data between clouds with its metadata intact.

That capabilities object is why CDMI is in this catalog beyond its history. The idea that a client should ask a server what it supports before using it, and get a structured answer, is the exact pattern the Model Context Protocol built its capability negotiation around fifteen years later. CDMI also shows the cost of being over-complete: an ISO standard that is technically excellent and nearly absent from the market it was written for. When I look at what a file surface for agents needs, CDMI is the reminder that discoverability of capabilities matters and that specifying everything is not the same as being adopted.