Glossary
Enterprise storage
Enterprise storage is the category of storage systems built to hold an organisation's shared, business-critical data, combining high availability, large capacity, data protection services and centralised management. The category is defined by service properties more than by technology: block, file and object systems, delivered as appliances or as software on standard servers, all fall within it.
Why enterprise storage decisions last
Storage is the layer every application, backup job and analytics pipeline depends on, and the hardest one to replace. Compute can be redeployed in hours; petabytes of data take months to migrate. Choices made about enterprise storage therefore persist for a decade or more, through several hardware generations, and determine how much the platform costs to grow, how it behaves when something fails and how many people it takes to run. For service providers and organisations with sovereignty requirements, the same decision also fixes where data physically sits and who is able to operate it.
Characteristics of enterprise storage
- Availability: redundant controllers or nodes, power and network paths, with upgrades and expansion carried out online. At 99.99% availability, the roughly 525,960 minutes in a year allow about 52.6 minutes of downtime.
- Data protection: RAID, erasure coding or replication inside the system, plus snapshots and replication to other systems or sites.
- Scale: hundreds of terabytes to many petabytes, shared by many applications.
- Data services: thin provisioning, compression and deduplication, encryption at rest, immutability, quotas and quality of service.
- Management: central administration, monitoring, automation APIs, access control and audit logging.
Enterprise storage architectures
| Architecture | Access unit | Typical protocols | Common uses |
|---|---|---|---|
| Storage area network (SAN) | Block (volume) | Fibre Channel, iSCSI, NVMe-oF | Databases, virtual machine disks |
| Network-attached storage (NAS) | File | NFS, SMB | Shared project files, home directories |
| Object storage | Object | S3 API over HTTP | Backup targets, archives, data lakes, media, AI data sets |
| Unified storage | Block and file in one system | Combination of the above | Consolidated general-purpose storage |
| Hyper-converged infrastructure | Block, pooled across compute nodes | Hypervisor-integrated | Virtualised application clusters |
Scale-up, scale-out and delivery models
Scale-up systems are built around a controller pair with shelves of drives. They grow by adding shelves until the controllers reach their limits, and a larger requirement then means a new system or a controller replacement. Scale-out systems are clusters of nodes, each adding capacity, CPU and network, and grow by adding nodes. New nodes join, data migrates onto them and old nodes retire, so hardware generations change without a wholesale migration.
Each architecture can be bought as an integrated appliance, licensed as software-defined storage on servers the customer chooses, consumed as capacity billed on premises, or rented from a public cloud. Many estates combine several, keeping latency-sensitive or regulated data on premises.
Data is also split by tier. Primary storage on flash holds live application data and is judged on latency; secondary storage holds backups, copies, file shares and analytics data and is judged on throughput and cost per terabyte; archive tiers hold long-retention data rarely read. Policies that move data between them as it ages are the subject of storage lifecycle management, and at petabyte scale the secondary and archive tiers usually hold most of the bytes.
What enterprise storage means for petabyte-scale operations
For a platform team managing petabytes across sites, the architecture choice shows up less on day one than in years three to ten, through growth, refresh, failure and cost.
- Most growth is unstructured. Backups, media, logs and AI data sets land on file and object platforms, while block arrays keep the latency-critical but smaller share of total bytes.
- The scaling model sets refresh pain. Scale-up estates face a migration every few years per array; scale-out platforms absorb new hardware and retire old nodes in the background.
- Hardware freedom affects cost at scale. Software-defined platforms let teams buy servers on their own cycle and pricing, while integrated appliances simplify support in exchange for vendor-specific hardware.
- Failure domains extend to sites. Light in fibre adds about 1 ms of round trip per 100 km, so synchronous protection suits metro distances and longer distances push toward asynchronous replication with a recovery point.
- Immutability has become a required data service, and its strength depends on the mode. Governance-mode S3 Object Lock can be bypassed by any identity holding
s3:BypassGovernanceRetentionthat sendsx-amz-bypass-governance-retention:true. Only compliance mode resists all users, including the account root, and only until the retain-until date. Object Lock requires versioning.
Scality enterprise storage
Scality RING is software-defined object and file storage on standard x86 servers, scaling to 300 billion objects in a single RING, with about 6 exabytes under management across the customer base. RING defines erasure coding schemes per storage class.
Scality ARTESCA is software-defined S3 object storage for backup, deployed on standard servers or as a hardware appliance and scaling to a validated 8.5 PB. Both products support S3 Object Lock in governance and compliance modes, with retention periods and legal holds.