Three letters decide how your servers see their storage: DAS, NAS or SAN. Choose wrong and you either overspend on a shared array a single server never needed, or you bolt a file appliance onto a problem that demanded shared block storage and spend years fighting it. The three are not competitors so much as answers to different questions — how the data is accessed, by how many hosts, and over what network. This guide explains what each architecture actually is, the block-versus-file distinction that underpins all of it, and a clear way to choose — including where an entry SAN like the HPE MSA or Dell PowerVault is the obvious sweet spot.
The three architectures, plainly
DAS (direct-attached storage) is disks that belong to one server — internal drives, or an external SAS enclosure cabled straight to that server's controller. There is no network in the path, so it is the simplest and lowest-latency option, but the capacity belongs to that one host. NAS (network-attached storage) is a file server on the network: clients mount shares over SMB or NFS and the appliance owns the filesystem. SAN (storage area network) presents raw block volumes to multiple servers over a dedicated storage network, and each server treats its volume like a local disk it can format and own.
The shorthand that resolves most confusion: DAS and SAN serve blocks, NAS serves files. A block device is a raw disk the host puts its own filesystem on; a file share is a filesystem the appliance manages and lends out. That single distinction drives almost every other difference in performance, sharing and cost below.
Block vs file: the distinction that matters
Block storage hands a server a volume of raw capacity. The server's operating system or hypervisor formats it (NTFS, VMFS, ext4, ZFS) and treats it as its own disk. That is exactly what virtual-machine datastores, databases and clustered applications expect, because they want low-latency, direct control of the storage and their own filesystem semantics. DAS delivers block locally; a SAN delivers block over a network so many servers can share one array.
File storage hands out a ready-made filesystem over SMB (Windows) or NFS (Unix/Linux). The appliance handles permissions, locking and the directory tree; clients just read and write files. That is ideal for user home directories, departmental shares, media libraries and any workload where many people or machines need the same files — but it is the wrong tool for a database or a VM datastore, which want a block device, not a file share.
- •DAS — block, one server, no network: internal disks or a direct SAS JBOD. Cheapest, fastest, not shared.
- •NAS — file, many clients, over the LAN: SMB/NFS shares. Great for files, wrong for VM/database block.
- •SAN — block, many servers, dedicated network: FC/iSCSI/SAS. Shared block for virtualisation and databases.
- •Many modern arrays are unified — one box serving both block (SAN) and file (NAS) — so the line blurs at the high end.
When to choose each
Choose DAS when a single server needs maximum performance per pound and the capacity does not need to be shared — a standalone database box, a backup target, or a branch server. Choose NAS when the requirement is shared files for people or applications: home drives, departmental shares, a media archive, or a backup landing zone that other systems write files to. Choose a SAN when several servers need to share block storage — most commonly a virtualisation cluster where hosts need shared datastores for live migration and failover, or a database tier that needs consistent low-latency block access.
The decision usually collapses to two questions. Is the access block or file? And is it used by one host or many? One host plus block is DAS; many clients plus file is NAS; many hosts plus block is a SAN. For the great majority of SMB and mid-market virtualisation projects that land on "many hosts, shared block," an entry SAN is the answer — and that is precisely the space the HPE MSA and Dell PowerVault occupy.
Performance, sharing and how each scales
DAS wins raw latency because nothing sits between the CPU and the disks, but it cannot be shared and scales only as far as the chassis and a JBOD or two. NAS scales as an appliance with expansion shelves and is limited at the top end by the file protocol and the LAN it rides on; it shares files beautifully but is not the right substrate for a hypervisor's datastores. A SAN scales furthest — dual-controller arrays grow to multiple petabytes across expansion enclosures and present shared block volumes to a whole cluster, which is why it underpins serious virtualisation and database estates. Our HPE MSA and Dell PowerVault pages cover how those entry SANs scale.
Resilience also differs. DAS resilience lives inside the one server (its RAID, its PSUs). NAS and SAN arrays are built around dual active-active controllers and redundant paths, so a controller or path failure does not take storage offline — the reason shared production storage almost always lives on an array rather than in a single host.
Cost, complexity and the entry-SAN sweet spot
DAS is cheapest and simplest; NAS adds a managed appliance; a SAN adds a storage network (Fibre Channel or iSCSI) and the array itself, which is more cost and more to run — but it buys shared block storage that a cluster genuinely needs. The good news is that the entry-SAN tier has made shared block storage affordable: an HPE MSA or Dell PowerVault ME5 gives you dual-controller, multi-path block storage for a virtualisation cluster at a fraction of historic SAN prices, with a choice of Fibre Channel, iSCSI or direct SAS connectivity.
If you have landed on "we need a SAN," the next decision is the fabric — iSCSI over Ethernet, Fibre Channel, or SAS direct-attach — which our iSCSI vs Fibre Channel guide covers. You can also build and price one in minutes with our HPE MSA or Dell PowerVault configurators, or, if you are replacing ageing kit, check its support runway with the storage end-of-life checker first.