NAS vs SAN vs DAS: Protocols, Workloads and How to Choose

NAS vs SAN vs DAS: Protocols, Workloads and How to Choose

NAS (network attached storage) serves files over NFS or SMB from a file system that lives on the storage system; a SAN (storage area network) serves block volumes over Fibre Channel, iSCSI or NVMe over Fabrics, and each host formats and owns the file system. DAS (direct attached storage) is a disk dedicated to one server, with no storage network. As of October 2026, vSphere 9.1, Hyper-V and Proxmox VE 9.2 document both NAS and SAN for VM storage, so the choice turns on sharing, multipathing, snapshots and protocol-specific features.

Related storage analysis is listed in Enterprise Techie’s topic index.

NAS vs SAN vs DAS at a glance

NASSANDAS
Access unitFiles and directoriesBlock volumes (SCSI LUNs, NVMe namespaces)Blocks on a local device
ProtocolsNFS, SMBFibre Channel, iSCSI, NVMe over FabricsSAS, SATA, NVMe over PCIe
File system lives onThe NASEach host (VMFS, NTFS or ReFS, for example)The one host
SharingMany clients; the server arbitrates locksMany hosts only with a cluster file system or cluster managerOne server
Typical networkStandard TCP/IP networkFibre Channel fabric (HBAs, switches), or Ethernet for iSCSI and NVMe/TCPNone; a cable to a controller or HBA
Typical workloadsFile shares, home directories, VDI profiles, NFS or SMB datastoresVMFS datastores, Hyper-V CSVs, clustered databases, hardened backup repositoriesSingle servers, HCI nodes, standalone backup repositories

What NAS, SAN and DAS are

The SNIA dictionary defines NAS as a storage device “that connects to a network and provides file access services to clients,” over protocols such as NFS and SMB. A SAN is a network “whose primary purpose is the transfer of data between computer systems and storage devices,” a term SNIA says is “usually (but not necessarily) identified with block services.” DAS is a storage device “locally connected to a server and dedicated to that server.”

The dividing line is who owns the file system. SNIA’s block storage entry says blocks are arranged into files “using a host file system,” so a SAN array tracks blocks while a NAS runs the file system itself. Object storage is a third model, covered in the object vs block vs file explainer.

Protocols and what each standard defines

ProtocolTypeDefined byWhat the standard covers
NFS 4.1 and 4.2FileIETF RFC 8881 (2020), RFC 7862 (2016)4.1: sessions, built-in locking, optional parallel NFS (pNFS). 4.2: server-side copy, sparse files, space reservations, Labeled NFS
SMB 3.xFileMicrosoft [MS-SMB2]Stateful file access; SMB 3.0 added per-share encryption, RDMA transports, failover with optional handle persistence, multiple connections per session
iSCSIBlockIETF RFC 7143 (2014)SCSI commands carried over one or more TCP connections that form a session
Fibre ChannelBlockINCITS Fibre Channel Technical Committee (formerly T11)Physical variants, framing and signaling, upper level protocol mappings, switch models
NVMe over FabricsBlockNVM Express Base and transport specificationsNVMe commands over RDMA and TCP; SNIA lists Fibre Channel as a further transport

SNIA’s dictionary calls SMB1 obsolete. NVM Express released its latest specifications on August 4, 2026, describes the set as NVMe 2.4, and labels the standalone NVMe-oF specification obsolete. SNIA notes each NVMe transport has “unique performance, availability, and scalability attributes”: speed depends on transport, network and array, not on the NAS or SAN label.

Multipathing

SNIA defines multipathing as a host directing requests to a storage device “on more than one access path.” On a SAN, the host does it: vSphere 9.1 documents path failover and load balancing across HBAs, switches and storage processors, and SNIA says most SANs are designed with redundant fabrics. The Proxmox multipath guide uses multipath-tools to merge per-path Linux devices that share one WWID and, for iSCSI, puts paths on separate NICs, networks and switches. RFC 7143 calls multiple NICs “simply paths to the same SCSI layer.”

File protocols build paths into the session. NFSv4.1 clients can bind many connections to one session (trunking); vSphere supports this for NFS 4.1 and offers no NFS 3 multipathing. SMB Multichannel aggregates bandwidth with fault tolerance across paths. Proxmox recommends a Linux bond for NFS and CIFS.

Shared access and locking

A LUN knows nothing about files, so hosts sharing one need a cluster file system. VMFS lets “multiple ESX hosts access the same VMFS datastore concurrently,” using on-disk locking. Windows Cluster Shared Volumes give cluster nodes simultaneous read-write access to one LUN. NetApp’s ONTAP docs say multiple clients can share a LUN “only if they are part of a cluster that prevents write collisions.” Fibre Channel zoning limits which nodes see each other.

On a NAS, the server arbitrates. RFC 8881 integrates locking into NFSv4 (share reservations, byte-range locks and delegations under one client lease), where NFSv3 used separate protocols such as NLM. SMB uses opportunistic locks and leases. vSphere uses client-side locking for NFS 3 and server-side locking for NFS 4.1, and never allows one datastore to be mounted with both.

Snapshots, replication and immutability

SNIA defines a snapshot as a “point in time copy of a defined collection of data,” such as files, LUNs or file systems. An array snapshot of the LUN behind a VMFS datastore therefore captures every VM on it. Synchronous replication commits at both sites before acknowledging the host; asynchronous replication acknowledges after the primary commits and forwards data later.

For ransomware, look for snapshots nobody can delete early. Vendor documentation, alphabetically: Dell lists “secure and immutable snapshots” for PowerStore; Everpure (formerly Pure Storage) lists SafeMode Snapshots and immutable snapshots for FlashArray; NetApp ONTAP can lock snapshots from 9.12.1, and its Autonomous Ransomware Protection covers NAS volumes and, from 9.17.1, SAN volumes. See storage-native ransomware detection.

Scale-out NAS and unified storage

Microsoft’s architecture guidance defines scaling up as changing a resource’s capacity and scaling out as adding instances. A scale-up NAS grows within one system; RFC 8881 notes that without pNFS, managing file access is “primarily the server’s responsibility.” Scale-out NAS spreads one namespace across nodes: Dell sells PowerScale as scale-out NAS, with pNFS in OneFS 9.15, and NetApp’s FlexGroup volume is “a scale-out NAS container.”

Unified storage serves file and block protocols from one system. Dell PowerStore lists NVMe-oF, Fibre Channel, iSCSI, SMB and NFS; Everpure says FlashArray “unifies high-performance block, file, and object storage”; ONTAP supports NFS, SMB, FC, FCoE, iSCSI, NVMe and S3. See Pure Storage vs NetApp.

How hypervisors consume NAS and SAN

VMware vSphere 9.1 supports VMFS, NFS, vSAN and vVol datastores. VMFS sits on block devices over Fibre Channel, FCoE, iSCSI or NVMe over Fibre Channel, RDMA (RoCE v2) or TCP; NFS datastores use NFS 3 or 4.1 and need no VMFS formatting. Broadcom deprecated vVols starting with VCF 9.0 and VVF 9.0 and plans to remove them in a future major release (KB 401070), so recheck array integrations built on them.

Hyper-V can keep VM configuration, VHDX disks and checkpoints on SMB 3 shares or on Cluster Shared Volumes carved from SAN LUNs; Microsoft advises NTFS for SAN-backed CSVs because ReFS there runs in redirected mode.

Proxmox VE 9.2 lists NFS and CIFS as shared file-level storage and iSCSI, Fibre Channel and SAS as shared block storage, with LVM splitting each LUN and multipathing “usually a requirement.” Shared storage lets running VMs live-migrate without copying disk images; see the Proxmox cluster guide.

iSCSI vs NFS for VM datastores

iSCSI (block)NFS (file)
What the host getsA LUN the host manages: VMFS, LVM, or NTFS for a CSVAn exported volume; vSphere mounts it without VMFS
MultipathingHost stack: ESX multipathing, Linux multipath-toolsNFS 4.1 session trunking; none for NFS 3 in vSphere; Linux bond in Proxmox
LockingVMFS on-disk lockingNFS 3 client-side, NFS 4.1 server-side (vSphere)
vSphere RDM and VM clusteringSupportedNot supported
vSphere NFS 4.1 limitNot applicableNo Site Recovery Manager array-based replication
Kubernetes in-tree access modesReadWriteOnce, ReadOnlyManyAdds ReadWriteMany
Proxmox VE snapshotsVolume chains on LVM, since Proxmox VE 9qcow2 snapshots

Sources: vSphere NFS datastore and storage comparison pages, Kubernetes persistent volumes, Proxmox VE storage.

Where each fits

  • Databases: SNIA says Fibre Channel SANs are often deployed for online transaction processing databases, and CSVs host SQL Server 2014 or later failover cluster instances. Microsoft also supports SQL Server database files on SMB 3 shares.
  • Home directories and shares: NAS. Windows Folder Redirection points known folders at “a directory on a file share.”
  • VDI: desktop disks live on VM datastores, while FSLogix profile containers “can be stored on file shares that support the SMB protocol,” per Microsoft.
  • Backup targets: Veeam accepts SMB and NFS shares as repositories, but its hardened repository “requires the block storage” (local, iSCSI or Fibre Channel) and rules out NFS and SMB. See all-flash backup target economics and the on-prem S3 guide.
  • Containers: the Container Storage Interface exposes “arbitrary storage systems” to Kubernetes; access modes depend on each driver, and volume snapshots require CSI.

HCI instead of an external SAN

Hyperconverged infrastructure pools servers’ own drives instead. vSAN aggregates local capacity devices “into a single datastore shared by all hosts in the vSAN cluster”; Storage Spaces Direct uses SMB 3 between nodes; Proxmox VE offers Ceph RBD. See what is HCI, vSAN alternatives and Nutanix external storage.

Evaluation checklist

  • Protocols and versions: NFS 4.1 or 4.2, SMB 3.x, iSCSI, Fibre Channel, NVMe/TCP or NVMe/FC, matched to what each hypervisor release supports.
  • Multipath: documented host configuration per operating system, plus NFS trunking or SMB Multichannel.
  • Snapshots and replication: granularity (file system, LUN or VM), synchronous and asynchronous modes, and hypervisor integration per protocol.
  • Ransomware protection: locked or immutable snapshots and detection, as the vendor documents them, and who can override a lock.
  • Capacity efficiency: thin provisioning and deduplication as documented, with any data reduction commitment in the contract. Proxmox warns that when thin storage runs full, every guest on it receives IO errors.
  • Hypervisor and Kubernetes support: compatibility listings for your hypervisor release, plus a CSI driver with the access modes and snapshots you need.

Frequently asked questions

What is the difference between NAS and SAN?

Where the file system lives. A NAS runs the file system and serves files over NFS or SMB to many clients. A SAN presents block volumes over Fibre Channel, iSCSI or NVMe over Fabrics, and each host formats them, so sharing a volume needs a cluster file system such as VMFS or CSV.

Is iSCSI a SAN or NAS?

SAN. RFC 7143 defines iSCSI as a transport for SCSI over TCP, so the host sees a LUN and formats it; VMware lists it as block access. It runs on ordinary Ethernet like NAS does, but NAS means file protocols such as NFS and SMB.

Is SAN faster than NAS?

There is no general answer. Results depend on transport, network design, multipathing and the array; SNIA notes each NVMe transport has its own performance attributes. Test with your own workload.

What is unified storage?

One system that serves both file protocols (NFS, SMB) and block protocols (Fibre Channel, iSCSI, NVMe over Fabrics). Dell PowerStore, Everpure FlashArray and NetApp ONTAP document this.

Sources

Last checked: October 2026. Guidance and product features change; check each source before you rely on it.