Glossary
Network-attached storage (NAS)
Network-attached storage (NAS) is file storage served over a network: a dedicated system holds a file system and shares its directories with clients through file protocols such as NFS and SMB.
Clients ask for files and directories by path. The NAS resolves each path, enforces permissions and turns the request into reads and writes on its own drives.
Why NAS matters at enterprise scale
File shares hold a large part of most organisations' unstructured data: engineering and design projects, media production, genomics and research output, home directories, and the staging areas of many analytics and AI pipelines. Applications written for a file system expect paths, locks and permissions, so NAS remains the default way to give many clients shared access to that data. Its limits show when file counts reach the billions and capacity grows into the petabytes.
How NAS serves files
A client mounts an NFS export or maps an SMB share, resolves a path one directory at a time, has its identity checked against the file's permissions, opens the file (often taking a lock or lease that governs caching), reads or writes byte ranges and closes it. Only the read or write step moves data. The rest are metadata operations, and on workloads with many small files they make up most of the work.
| Protocol | Typical clients | Notable capabilities |
|---|---|---|
| NFS version 3 | Unix and Linux | Simple, widely deployed, stateless server model |
| NFS version 4.1 | Unix and Linux | Sessions, delegations, and parallel NFS (pNFS), which lets clients read and write data directly on several storage servers |
| SMB 3 | Windows, also macOS and Linux | Leases, multichannel across several connections, per-share encryption |
Permissions follow two models. POSIX mode bits grant read, write and execute to owner, group and others; access control lists in NFSv4 and Windows grant rights to any number of users and groups with inheritance down the tree. Shares served over both protocols map Unix IDs to Windows identities so one set of permissions applies either way.
Scale-up and scale-out NAS
A scale-up NAS, or filer, has one or two controllers serving file systems from attached drive shelves. Each file system sits behind one controller at a time, so its performance is bounded by that controller, and capacity stops at the controller's drive limit.
A scale-out NAS is a cluster of nodes presenting one namespace. Files and metadata are spread across nodes, clients connect to any node, and adding nodes adds capacity and performance together. Internally it is a distributed file system exposed through NFS and SMB.
Protection works at two levels. Within the system, drives sit under RAID or erasure coding and snapshots record point-in-time images that users can browse to recover earlier file versions. Beyond it, file-level replication copies changes to a second NAS, and backup software copies data to a separate target, sometimes streamed directly by the NAS through the Network Data Management Protocol (NDMP).
NAS and object storage
NAS organises data as a directory hierarchy with in-place updates to byte ranges, file locking and POSIX-style permissions. Object storage organises data as a flat namespace of keys in buckets, written as whole objects over HTTP, with rich per-object metadata and no tree to traverse. NAS fits applications written for file systems; object storage fits applications written for the S3 API and data sets counted in billions of items. The object storage vs NAS entry sets out the comparison in full.
What NAS means for petabyte-scale file data
For teams holding petabytes of file data, the pressure points are file count, scanning and silos more than raw capacity.
- Small files cost time more than space. At four round trips of 0.5 ms per file, writing a million 4 KB files takes about 33 minutes, while the same 4 GB written as one file at 1 GB/s takes about 4 seconds.
- Backups of large file systems are dominated by scanning. Walking hundreds of millions of files to find what changed can take longer than copying the changes, so backup windows stretch as file counts grow even when daily change is small.
- Scale-up filers multiply into silos. Each has its own namespace, controller limit and refresh cycle, so growth adds systems, mount points and migrations.
- Newer consumers often speak S3. Analytics engines, AI frameworks and cloud-native applications read through the S3 API, so file data that has to feed them is either copied or exposed through both file and object interfaces.
Scality and file storage
Scality RING is software-defined object and file storage on standard x86 servers, so applications written for file systems and applications written for the S3 API can share one platform. A single RING scales to 300 billion objects. For very small items that count is the bound reached first: 300 billion items of 4 KB total 1.2 PB, which puts object count ahead of capacity in planning for small-file estates.