Object Storage
Object storage keeps data as immutable objects with metadata in a flat namespace, scaling to enormous capacity with simple access.
Data as objects, not files
Object storage manages data as objects: each is a blob of bytes plus metadata and a unique key, stored in a flat namespace called a bucket. Unlike a filesystem with nested directories, there are no real folders; the slashes in a key are just part of its name. This simple model scales to trillions of objects and vast total capacity.
Key properties
- Flat namespace: objects addressed by key within a bucket.
- Rich metadata: user- and system-defined attributes travel with each object.
- Immutability: objects are typically replaced whole, not edited in place.
- Access over HTTP: a simple API rather than a filesystem mount.
- Durability: automatic replication or erasure coding across devices.
Durability and erasure coding
Object stores achieve very high durability by spreading each object's data across many devices with redundancy. Erasure coding splits an object into fragments plus parity, so the original survives the loss of several fragments while using less space than full replication. This is how a store can promise that stored data is extremely unlikely to be lost.
Immutability suits science
Because objects are written whole and replaced rather than edited, object storage naturally supports the immutable-raw-data discipline: a measurement is written once, keyed, and never altered. Corrections become new objects. This aligns with content addressing and versioning.
Foundation of lakes
Object storage is the usual foundation for a data lake and a lakehouse, holding raw diagnostic and simulation files, curated Parquet tables, and the deposited published record. Its scale, durability, and metadata make it well suited to the long-lived archive a fusion research program needs.