Glossary
Multi-protocol storage
Multi-protocol storage stores one copy of data and presents it through several access protocols, typically file shares and the S3 API, so different applications can work on the same dataset without copying it between systems. The system translates names, paths and permissions between two models that were not designed to match.
One copy instead of several
A genomics lab shows the problem it solves. Sequencers write results to a file share. The analysis pipeline reads its input through S3. Without a shared copy, a nightly job moves data from the share to a bucket, two copies are paid for and protected separately, and anything written after the copy runs is invisible to the pipeline until the next night. With multi-protocol access, the sequencers write files and the pipeline reads the same data as objects minutes later. One protection scheme covers it, and there is no copy job to fail. The same shape appears in many an AI data pipeline, where capture systems write files and training tools expect S3.
Where do file and object behaviour clash?
| Behaviour | Through a file share | Through S3 |
|---|---|---|
| Renaming a folder | One quick metadata change | Every object under the prefix is copied to a new key |
| Upper and lower case | Ignored by Windows clients | Keys are case sensitive |
| Permissions | Detailed per-user and per-group access lists | Bucket and user policies, which carry less detail |
| Locks | Stop a second writer | Do not stop an S3 overwrite |
| Partial updates | Part of a file can be rewritten | The whole object is replaced |
None of these is a defect. Each comes from how the two models were built, and each matters only when both kinds of client write to the same data.
A writer of record per dataset
Shared data stays predictable when each dataset has one protocol that writes and others that read. Files ingested over a share and read through S3 give every reader one consistent view. When both sides write, a folder rename on one side or a lock the other side ignores produces errors that are slow to trace. Permissions follow the same principle: the system maps file access lists to S3 policies, but the mapping flattens detail, so one model is treated as authoritative for each dataset. Operations that cross the seam, such as listing a very large folder through S3, are where translation costs show, and they are worth timing before a pipeline depends on them.
How it differs from unified storage and gateways
Unified storage serves block and file from one array, usually as separate volumes and shares. Multi-protocol storage is about one dataset seen through several doors. A cloud storage gateway is different again: it presents a file or block interface locally and stores the data in a cloud object service. The file side behaves as a NAS share, and the object side as S3-compatible storage.