Glossary

Object storage

Object storage keeps data as self-contained objects, each holding the data, its metadata and a unique key, in one flat pool reached over the network. Objects are written and replaced whole, never edited in place, and capacity grows by adding servers rather than by deepening a folder tree.

How does object storage work?

An application sends an object, such as a scan, a video or a backup file, and receives a key in return. The system decides which servers and drives hold it. To read it back, the application presents the key over HTTP, most often with the S3 API. Objects sit side by side in flat buckets, with no nested tree to walk, so a bucket with billions of items is looked up as quickly as one with ten. Metadata travels with each object, so labels for governance, retention or search are stored with the data they describe.

The pool grows by adding servers. Protection usually comes from erasure coding, which splits an object into data and recovery pieces spread across the cluster and needs less raw capacity than keeping three full copies.

What is the rule that decides whether data is object data?

Objects are replaced whole. An application that changes the middle of a file would have to rewrite the entire object each time, so if a program rewrites parts of its files, the data is not object data. The test is the write pattern, not the file type or size.

What workloads is object storage a good fit for?

  • Backups and recovery copies, especially with retention locks.
  • Media libraries and medical image archives.
  • Research datasets and AI training data, which are written once and read many times.
  • Long-term compliance records.

It is a poor fit for databases that update blocks, virtual machine disks and shared folders where people edit files all day. Those belong on block storage or NAS. Many organizations run both, and the comparison is covered in object storage vs. NAS.

How does it stay safe and keep growing?

Because pieces are spread across many servers, the loss of a drive or a whole server does not lose objects, and capacity grows without migration. Locking is possible at the object level: S3 Object Lock keeps an object unchanged until its retention date, which is why object stores are common backup targets. Systems that follow the S3 API are called S3-compatible storage.

A last point is how objects are found. With no tree to walk, applications keep their own index of keys, or use metadata queries and listings by prefix. Teams moving from file shares often find this the larger adjustment, because folder names carry meaning that now has to live in the key or in metadata.

Frequently asked questions

Is object storage only for archives?

No. Archives are the best-known use, but analytics, AI pipelines and backup repositories all read and write it continuously. The restriction is on rewriting objects, not on how often they are used.

Does object storage need a file system?

No. It has no folders. Tools show a folder-like view by treating slashes in the key as separators, but the namespace underneath is flat.

How is object storage priced to scale?

By adding servers and drives. Capacity and throughput grow together, so adding capacity does not usually require replacing the whole system.