Fibre Channel part 4: redundancy and multipathing

Closes the FC series after fabric login.

Two fabrics, never joined

Storage access is mission critical, so no single point of failure anywhere. Build two complete FC networks: Fabric A and Fabric B.

Every server and every storage controller connects to both fabrics via redundant HBA ports (S1-A into A, S1-B into B).

The fabrics are not cross-connected. Switches share state (Domain IDs, FCNS, zoning), so a fault propagating between fabrics would take down both. Hosts touch both; switches never do.

With dual controllers on the array, a single server ends up with four paths to its LUN (2 fabrics x 2 controllers). Zoning is configured per fabric; LUN masking on the array covers all the ports.

Paths and picking between them

The Target Portal Group is all the storage ports an initiator may reach its LUN through (CTRL1-A, CTRL1-B, CTRL2-A, CTRL2-B).

ALUA (Asymmetric Logical Unit Access) is how the array tells the client which paths to prefer: direct paths to the controller that owns the LUN are optimized, paths via the partner controller are non-optimized.

Multipathing software on the initiator picks the path or paths: active/active or active/standby, with automatic failover when a path dies. Every mainstream OS ships it.

The key mental model shift from Ethernet: in IP networking the network devices make the forwarding decisions; in SAN the path intelligence lives on the client. FC initiators discover their paths automatically through FLOGI/PLOGI/PRLI, then multipathing chooses.