Glossary
Cloud storage
Cloud storage is data storage delivered as a network service: capacity is requested through an API or console, provisioned on demand from a shared pool and billed or accounted for by use. The infrastructure is run by a public provider, by the organization itself as a private cloud, or by a combination of the two.
The term follows the NIST definition of cloud computing (SP 800-145): on-demand network access to a shared pool of configurable resources, storage among them, that can be provisioned and released rapidly with minimal management effort.
Why cloud storage matters for enterprise data strategy
The cloud storage model changed expectations inside enterprises as much as it changed where data sits. Application teams now expect storage to appear by API call in minutes, grow without a procurement cycle and be charged to the project that uses it. Platform teams running on-premises storage are measured against that experience, while teams in public cloud find that at petabyte scale the meters for requests, retrieval and data transfer can outweigh capacity on the bill.
Characteristics and types of cloud storage
NIST lists five essential characteristics of cloud services, each with a storage form.
- On-demand self-service: buckets, volumes and shares created by API call without an operator.
- Broad network access: data reached over standard protocols such as HTTPS, NFS, SMB and iSCSI.
- Resource pooling: many tenants share the same drives and servers, separated logically.
- Rapid elasticity: capacity grows and shrinks with use.
- Measured service: capacity, requests and transfer metered per account.
Cloud services expose the same three storage models as on-premises systems. Object storage holds data as objects in buckets addressed by key over HTTP and is the most common form; the S3 API is implemented by many other services and products, described as S3-compatible storage. Block storage provides fixed-size volumes attached to virtual machines, and file storage provides managed NFS or SMB shares.
Deployment models
| Model | Who operates the infrastructure | Who uses it |
|---|---|---|
| Public cloud | A provider | Any customer of the provider |
| Private cloud | The organization or a contractor on its behalf | One organization |
| Community cloud | Members of a community or a third party | A group with shared requirements |
| Hybrid cloud | A combination of the above | Data and applications portable between them |
A private cloud storage service offers the self-service API model on infrastructure the organization controls. A hybrid cloud links private and public storage by replication or tiering, and an estate across several public providers is multi-cloud.
Pricing dimensions
- Capacity: gigabytes stored per month, by storage class.
- Requests: charges per thousand writes, lists or reads.
- Retrieval: charges per gigabyte read from infrequent-access and archive classes.
- Data transfer: charges for data leaving the provider or crossing regions, described under egress fees.
- Minimums: charges for a class's minimum duration or object size, whether or not the data meets it.
Total cost depends on access pattern as much as volume. 100 TB read once a year produces a bill dominated by capacity; the same 100 TB read daily from outside the provider produces one dominated by transfer. Request charges weigh most on workloads with many small objects, such as logs, thumbnails or AI datasets split into millions of samples, where per-request fees can exceed capacity charges for the same data.
What cloud storage means for petabyte-scale estates
At large scale the decision is rarely all-cloud or no-cloud. It is a placement decision per dataset. Data with unpredictable demand, a short life or global users benefits from public cloud elasticity. Data that is large, long-lived and read often, such as AI training corpora, media libraries, backup repositories and analytics lakes, accumulates capacity and transfer charges month after month, and moving it out later incurs egress charges on every byte. Data gravity makes the placement stickier with each petabyte added.
Sovereignty adds a second axis. Laws and contracts that restrict where data is stored, and which jurisdictions can compel access to it, push some datasets toward private or sovereign cloud regardless of price. The usual outcome in large enterprises is a hybrid estate in which the same API, most often S3, runs in public regions and on infrastructure the organization controls, so applications move between them without rewriting.
Operating model matters as much as location. A private cloud storage service delivers the self-service experience only when provisioning, quotas, chargeback and lifecycle are handled by storage automation. Without that, application teams route around it to the public cloud and the estate fragments across both.
Cloud storage on Scality platforms
RING is software-defined object and file storage on standard x86 servers, with S3, NFS and SMB access and a unified S3 namespace across sites and clouds, which brings the S3 model of public services onto infrastructure an organization runs itself. Scality's ADI deploys on-premises, air-gapped or in sovereign cloud, with data residency enforced by policy at the namespace level.