Glossary
Storage area network (SAN)
A storage area network (SAN) is a dedicated network that gives servers block-level access to shared storage, so that a volume on a remote array appears to the server's operating system as a locally attached disk.
The server formats the volume with its own file system and manages it as its own. The SAN carries SCSI or NVMe commands between server and array.
Why SANs matter in enterprise infrastructure
SANs carry the workloads most organisations treat as their core: databases, ERP systems and the virtual machine disks of virtualised clusters. Their strength is consistent low latency for small block operations. Their weakness at petabyte scale is cost and growth, since arrays and Fibre Channel fabrics are priced for performance and much of the data that accumulates on them over the years never needed block access.
How block access over a SAN works
A SAN presents storage as numbered blocks on a logical unit, identified by a logical unit number (LUN) or, with NVMe, a namespace. The server is the initiator and the array is the target. The initiator reads and writes ranges of blocks; the array knows nothing about files, which exist only in the file system the server builds on the volume. This is the block storage model of a local disk extended over a network, and because each file system assumes sole control of its volume, two servers share a LUN only through a cluster file system or cluster-aware application.
Two controls decide which server sees which volume. Zoning, set on the switches, lists the ports allowed to talk to each other. LUN masking, set on the array, lists which initiators may access each LUN. Together they stop a server from discovering, and possibly formatting, a volume that belongs to another.
Enterprise SANs are commonly built as two independent fabrics. Each server and each array controller connects to both, and multipath software spreads I/O across the paths and moves traffic off a failed one, so losing an entire fabric leaves half the paths in service.
SAN protocols
- Fibre Channel (FC): a transport built for storage, on its own switches and adapters. It is lossless, because a sender transmits only when the receiver has advertised free buffer space, which keeps latency consistent under load.
- iSCSI: SCSI commands over TCP/IP on standard Ethernet. TCP handles delivery, so packet loss leads to retransmission, and protocol processing adds host CPU work unless offloaded.
- FCoE: Fibre Channel frames carried in Ethernet configured for lossless operation.
- NVMe over Fabrics (NVMe-oF): the NVMe command set, with its many parallel queues, over Fibre Channel, RDMA or TCP, removing the SCSI translation layer.
Whatever the protocol, small-block workloads usually reach the array's controller and media limits before the link limit. Queue depth per LUN and per port, controller CPU and cache hit ratio set the IOPS a volume sustains, which ties SAN performance to the array more than to the network link.
SAN, NAS and object storage compared
| Property | SAN | NAS | Object storage |
|---|---|---|---|
| Unit of access | Block | File | Object |
| File system location | On the server | On the NAS | None; flat namespace of keys |
| Protocols | FC, iSCSI, FCoE, NVMe-oF | NFS, SMB | S3 API over HTTP |
| Sharing between servers | Requires a cluster file system | Native | Native |
| Scaling model | Array or cluster of arrays | Filer or scale-out cluster | Scale-out cluster |
The network-attached storage entry covers file access in more detail.
What SANs mean for large-scale storage estates
In an estate that holds several petabytes, the SAN usually remains the right home for a small, latency-critical share of the data and an expensive home for much of the rest.
- Block arrays grow in array-sized steps. Each array is its own island of capacity, management and refresh cycle, and each refresh means migrating volumes off the old system.
- Fibre Channel is a separate network with its own switches, optics, skills and support contracts. iSCSI and NVMe over TCP run on Ethernet the team already operates, at the cost of tuning that network for storage traffic.
- Long-lived SANs collect file shares, backup repositories and archives presented as volumes, paying block-tier prices for data that is rarely read. Moving that data to file or object platforms is the purpose of SAN consolidation.
- Block storage has no shared namespace. Data on a LUN is reachable only through the server that owns its file system, which makes it awkward input for analytics and AI pipelines where many clients read the same data at once.
Scality and SAN environments
Scality RING is software-defined object and file storage on standard x86 servers, reached through the S3 API and file interfaces over ordinary IP networks, with no Fibre Channel fabric involved. In estates that keep databases and virtual machine disks on block arrays, RING is the kind of platform that takes over unstructured data, backups and archives, and a single RING scales to 300 billion objects.