Multipath (dm-multipath) presents two or more physical I/O paths to the same disk as a single device, so storage survives a cable, switch, or adapter failure.

In enterprise storage the disk often lives in a separate array reached over a network (a SAN). If there's only one route — one cable, one switch, one host adapter — that route is a single point of failure. Multipath wires up multiple independent routes to the same disk and then hides them behind one logical device:

Server ──→ HBA1 ──→ Switch1 ──→ Storage
       └─→ HBA2 ──→ Switch2 ──┘

What it provides:

• Failover — if one path dies (pulled cable, dead switch), I/O silently continues on another.
• Load balancing — spread I/O across paths for more throughput.
• High availability — no single component can take the disk offline.

The combined device appears at /dev/mapper/<name>, and everything above it (partitions, LVM, the filesystem) just uses that one path-independent device. It's mainly seen with SAN, iSCSI, and FCoE storage in data centers. For network-attached storage you can get similar resilience at the network layer with NIC bonding.

Go deeper:

• Multipath I/O — Wikipedia — redundant I/O paths, failover, and load balancing for SAN/iSCSI storage.