Glossary
Block storage
Block storage presents capacity to a server as a raw volume made of fixed-size, numbered blocks, which the server's operating system formats and uses as its own disk. The storage system tracks only block addresses, never files, so small in-place updates are fast while rules that depend on file content are out of reach.
A volume is only numbered blocks
The array hands a server a volume of a set size. Everything above the block level, the file system, the database files and the virtual machine images, belongs to the server. The array receives requests to read or overwrite particular blocks and has no knowledge of what they contain. Volumes usually travel over a storage area network, using Fibre Channel, iSCSI over Ethernet or NVMe over Fabrics, and servers treat them as local drives.
One database update, traced
An order system changes the delivery address on one order. The database works out which 8 KB page holds that row and asks the operating system to write it. The operating system turns the page into a write to a block address on the volume, and the array overwrites those blocks where they sit. Nothing else on the volume moves, no file name appears anywhere in the exchange, and the database receives its acknowledgement quickly. Thousands of such updates a second, scattered across the volume, is the pattern block storage is built for, and it is the same pattern a virtual machine disk produces.
Where does block storage stop being the right tool?
Because the array cannot see files, it cannot keep one document for seven years, expire one record after a month or tag data for search. A volume normally has a single writer, and two servers mounting it with an ordinary file system will corrupt it unless a cluster-aware layer coordinates them. Recovery is coarse in the same way: a volume snapshot restores the whole volume, and retrieving one deleted file means mounting an older copy and copying it out by hand. Cost is the last limit. Block capacity is priced for low latency, and backups, archives and media that settle there over the years are often cheaper to keep on object storage.
What file and object add on top
- File storage adds folders, permissions and locking so many users can share data, as in network-attached storage.
- Object storage adds per-object metadata and a flat namespace that grows by adding servers.
- Block storage adds nothing above the disk, which is what keeps its response times short and predictable.