Cloud Parts

Object, block or file storage?

DimensionObject storageBlock storageFile storage
You access it throughAn HTTP API, by keyA raw disk mounted on one machineA shared folder over NFS or SMB
Has a file systemNo. Keys only look like folders; a mount tool can add one on topNo. You format one onto it (ext4, NTFS) before useYes, that’s the product
Shared between machinesYes, by anyone with permissionUsually one VM at a timeYes, many at once
Edit part of a fileNo; rewrite the whole objectYes, block by blockYes
Rename a folderCopy and delete every object under the prefixInstant, via the file system on itInstant
Typical latencyTens of millisecondsUnder a millisecond to a fewA few milliseconds
ScalePractically unlimitedTens of TB per volumeGrows automatically to petabytes
Reach for it forMedia, backups, logs, data lakesDatabases and boot disksShared content, legacy apps, home folders

Rule of thumb

A typical app uses all three: block storage under its database, object storage for uploads and backups, and file storage only when several servers truly need the same folder. Software that expects files can still use a bucket through a mount such as Amazon S3 Files or Mountpoint, with limits.

Sources