AI Glossary / Data engineering and architecture / Advanced
Storage area network
Definition
A storage area network, or SAN, is a network primarily used to transfer data between computers and storage devices, or among storage devices. It is commonly associated with block-level storage access, where a host works with storage volumes rather than requesting named files from a file server. A SAN includes connectivity and management arrangements; it is not simply another name for a disk array.
What it means
Storage need not sit inside the computer that uses it. A SAN connects hosts to storage through a network designed and managed for that purpose. To the host, a presented storage device can behave much like locally attached storage even though the path includes network infrastructure. The distinction matters when planning capacity, access and failure handling. A server's ability to see a volume does not mean every other server should access it, and shared access does not automatically coordinate simultaneous writes. The application and filesystem arrangements must match the intended use.
How it works
Hosts communicate with storage through configured paths, which may include adapters, switches and storage controllers. Storage management determines what is presented to each host and how connections are organized. SANs can use different technologies, including Fibre Channel and Ethernet-based approaches; the term does not require Fibre Channel. Designs may provide multiple paths for resilience, but their behavior depends on the configured hardware and software. The usual contrast with network-attached storage is the access model: block-oriented access is commonly associated with SANs, while NAS provides file access services. Product boundaries can overlap.
Three practical examples
- A database infrastructure team provisions a storage volume to an authorized host and measures the latency of its actual workload.
- A virtual-machine platform uses centrally managed storage while testing how host access behaves when a configured path fails.
- An AI data-processing service evaluates whether its storage network can sustain concurrent dataset reads and checkpoint writes without unacceptable waiting.
Think of it this way
Imagine several workshops connected to a central materials store by managed delivery routes. Each workshop has permission to use specified supplies, and alternative routes may be available. More routes can improve access, but they do not duplicate the supplies or ensure that two workers update the same inventory correctly. The analogy separates connectivity, capacity and coordination, which are related but different concerns in a storage system.
Worked example
Consider an original simplified system with two independent active data paths, each capable of carrying 400 megabytes per second for the workload being modeled. Assume the host, storage and scheduling can use both paths fully. A 600-megabyte-per-second workload fits within their combined 800-megabyte-per-second capacity. If one path fails, only 400 remains, leaving a 200-megabyte-per-second shortfall while the workload continues at the same rate. Redundant connectivity therefore does not necessarily preserve full performance after a failure. These figures are invented capacity assumptions, not advertised interface speeds or a benchmark. They also do not imply that every multipath configuration combines bandwidth in this way; some configurations use a path mainly for failover or have other bottlenecks.
Illustrative case study
In a fictional analytics operation, an infrastructure review records two routes from a processing host to storage and labels the service fully resilient. During a planned test, the remaining route keeps the volume reachable but the processing queue grows. The team revises its acceptance criteria to include degraded-mode throughput and completion deadlines, rather than checking reachability alone. It also tests recovery separately from path switching. The scenario illustrates why a diagram with duplicate connections is insufficient evidence of application continuity. It does not report a real failure or prescribe a particular SAN vendor, protocol or configuration.
Common misconceptions and mistakes
A SAN is not necessarily Fibre Channel, and it is not the same thing as a shared folder. Multiple paths also do not create additional copies of the stored data. Another mistake is treating shared block access as permission for unrelated hosts to mount and modify an ordinary filesystem concurrently; coordination requirements still apply. Finally, storage capacity, network bandwidth and operation latency are different measures. Having enough space for a dataset does not establish that the application can read it quickly enough, and a high nominal link speed does not prove end-to-end performance.
Business use
The concept helps teams assess infrastructure used by databases, virtualization platforms and data-intensive applications. Define the workload in terms of read and write patterns, expected concurrency and required response times, rather than storage size alone. For AI workloads, distinguish dataset loading from periodic checkpoint traffic and test their overlap when relevant. Clarify the responsibilities for access configuration, monitoring and recovery. Evaluate normal and degraded operation against business deadlines. A SAN can be part of an availability design, but its name does not establish backup quality, application consistency or the amount of disruption a failure will cause.
People also ask
Must a SAN use Fibre Channel?
No. The term covers a storage-network purpose, not a single transport technology.
Is a SAN the same as NAS?
No. SAN commonly refers to block-oriented storage networking, while NAS provides file services, although systems can offer both.
Do two paths guarantee unchanged performance after one fails?
No. Remaining capacity and the application's requirements determine the effect.
Check your understanding
What combined capacity did the toy model assume?
Answer: Eight hundred megabytes per second.
What remained after one path failed?
Answer: Four hundred.
Why did a 600-megabyte-per-second workload develop a shortfall?
Answer: Its demand exceeded the surviving path's assumed capacity by 200.
What did the fictional team add to its test?
Answer: Performance and completion criteria during degraded operation, rather than reachability alone.
Related terms
- Network-attached storage
- block storage
- Fibre Channel
- iSCSI
- multipathing
- storage volume
Learn next
Draw a host, two paths and one storage system. Mark which components are actually independent and which are shared. Assign hypothetical workload demand and surviving-path capacity, then calculate whether a deadline could still be met after a failure. Separately explain what would protect the data itself, since alternate connectivity does not provide a backup.
More in Data engineering and architecture
- Spindling - Spindling is the placement of different files on different physical hard disks to distribute storage activity and reduce competition for the same device.
- Spurious Tuple - A spurious tuple is an extra tuple produced when relations are joined in a way that reconstructs combinations not present in the original relation being...
- Staging (data) - Data staging is the use of an intermediate location to hold data between its source and a later processing or delivery step.
- Storage model - A storage model describes how data is physically organized in a data store, including how record attributes are grouped and represented for access.
- Stream processing - Stream processing is the continuous or incremental computation of results from a flow of incoming events or records.
- Streaming analytics - Streaming analytics is the ongoing analysis of incoming data streams to produce timely metrics, patterns, alerts, or other analytical results.
Sources
SNIA, What Is a Storage Area Network (SAN)?: https://www.snia.org/education/storage_networking_primer/san/what_san SNIA Japan storage-networking dictionary, storage area network, including its English definition: https://www.snia-j.org/dictionary/1568/ SNIA, Network Attached Storage: https://www.snia.org/education/online-dictionary/term/network-attached-storage The bandwidth calculation, workshop analogy and analytics scenario are original educational illustrations. The calculation explicitly assumes simultaneous use of independent paths and should not be treated as the default behavior of every SAN or multipath configuration.
Previous: Stopping time Next: Storage model