Glossary
Object storage
Object storage keeps data as self-contained objects, each made of the data, a description of it and a unique name, in one flat pool reached over the network.
How object storage works
An application hands over a scan, a video or a backup file and gets back a name, called a key. The storage system decides which servers and drives hold it, and the key is all it takes to get it back. To fetch it, the application presents the key through a set of web-style requests, most often the S3 API.
Objects live in containers called buckets rather than folders. There is no folder tree to walk or maintain, which lets the pool grow to very large numbers of objects.
An object, its key and its bucket
A file system finds data by walking folders. An object store finds it by name, in one flat pool.
Object storage workloads
Object storage suits data that is written once and read many times:
- Backups and recovery copies.
- Media libraries, medical images and video archives.
- Research datasets, logs and AI training data.
- Records kept for years for compliance.
It is a poor fit for data that changes a few bytes at a time:
- Databases and virtual machine disks, which need block storage.
- Shared folders where people open and edit files in place. The comparison with NAS covers where that line falls.
The dividing line is how changes are written. When an object changes, the whole object is replaced, often as a new version. Some platforms also let applications that expect files reach the same data over file protocols, an approach called multi-protocol storage.
Object storage scaling and protection
Object stores are usually built from many ordinary servers working as one system, an approach called scale-out storage. Capacity grows by adding servers, and the pool stays one system to run.
Protection is spread across those servers too. Many platforms use erasure coding, which splits each object into pieces plus spare recovery pieces. A failed drive or server then loses nothing, within the number of failures the scheme is designed to survive, and the overhead is lower than keeping three full copies.
Each object can also carry labels, such as a patient ID or a camera name. Those labels make very large collections easier to manage, search through an index and govern.
Scality RING and object storage
Scality RING is software-defined object storage that runs on standard x86 servers and serves data over the S3 API, with erasure coding set per storage class and S3 Object Lock in governance and compliance modes. Scality customers run about 6 exabytes on RING in total.