Local storage vs. network-attached storage: what actually differs
Kort antwoord
Local storage physically lives on the same host as the server, so it's fast because there's no network hop involved, but it's tied to that specific machine. Network-attached storage, block storage volumes, is detachable and can move between servers, at the cost of a network hop's worth of latency. From inside the operating system both can look identical, a mounted disk is a mounted disk, until a host failure, a migration, or a resize forces the difference to matter.
Same mount point, different physical reality
Local (instance) storage sits inside, or directly attached to, the physical host running your server. There's no network between the disk and the CPU reading from it, which is exactly why local storage is fast: every read and write travels the shortest possible path. The trade-off is that local storage belongs to that one physical machine. If the server is rebuilt on different hardware, or the underlying host fails, local storage doesn't travel with it. It stays behind.
Network-attached storage takes the opposite trade. A block storage volume lives on separate storage infrastructure and reaches your server over the network, so every read and write crosses that network hop, adding latency that local storage simply doesn't have. What you get in return is detachability: the volume isn't bound to one physical host, so it can be moved to a different server, reattached after a rebuild, or resized independently of the machine using it.
Why this trips people up
From inside the operating system, the distinction mostly disappears. A mounted disk shows up as a mounted disk, whether it's a local NVMe drive bolted to the host or a network-attached volume reached over the network. Nothing in a normal df -h or Disk Management view forces you to notice which one you're looking at, and for day-to-day reads and writes, both just work.
The difference surfaces the moment something happens to the underlying hardware. A host failure that would otherwise mean rebuilding a server from a backup is a non-event for data sitting on a network-attached volume: reattach it to a new host and carry on. The same failure against local storage means the data is gone unless it was backed up separately, because local storage never had anywhere else to be. Migrations and resizes tell the same story: moving a server to different hardware, or growing a volume beyond what its host has room for, is straightforward with network-attached storage and effectively impossible with local storage without a full data copy first.
Choosing between them
| Factor | Local storage | Network-attached storage |
|---|---|---|
| Latency | Lowest, no network hop | Slightly higher, travels over the network |
| Tied to physical host | Yes | No, can be detached and reattached |
| Survives a host failure | No, unless backed up separately | Yes, the volume itself isn't affected |
| Best fit | Latency-sensitive workloads where the server and its data can be treated as one unit | Anything you might need to migrate, resize, or recover independently of the host |
Neither option is universally right. Fast, latency-critical workloads that can tolerate rebuilding from a backup lean toward local storage. Anything where the data needs to outlive the specific piece of hardware it started on, or where you expect to resize or migrate later, leans toward network-attached storage instead.