iSCSI Multipath and SAN Troubleshooting Playbook

Two NICs are not automatically two paths

True multipath design requires independent usable network paths and host multipath software that understands them. If both NICs depend on one switch, one VLAN failure or one storage port, the design still contains a shared failure point.

Record the path

• Host initiator IQN.
• Host NIC/IP for each path.
• Storage target IQN.
• Target portal IPs.
• VLAN and switch port for each path.
• LUN identifier.
• Multipath policy.

MTU

If jumbo frames are used, every hop in the storage path must support the selected MTU. Partial jumbo configuration can create intermittent or confusing behaviour. Test end to end with a packet size that proves the intended MTU rather than checking only the server NIC setting.

Latency isolation

NetApp documents cases where iSCSI workloads show low array-side latency but high network delay. That is why the installer should check host, network and storage metrics independently before blaming disks.

Network checks

show interfaces counters errors
show interfaces <port>
show mac address-table

Failover test

1. Confirm all expected paths are active.
2. Generate controlled storage I/O.
3. Disable one approved network path.
4. Confirm the host continues I/O through another path.
5. Restore and confirm the path rejoins.

What does not work

• Using ordinary NIC teaming where the storage vendor requires multipath I/O.
• Setting jumbo frames on only host and array.
• Allowing iSCSI to share an oversubscribed user uplink without measuring contention.
• Changing path policy while troubleshooting an unrelated array issue.

Session vs path

An iSCSI target can be reachable while multipathing is still incomplete. Confirm every intended session/path is established and that the host multipath layer presents one logical device rather than duplicate unmanaged disks.

Switch design

Keep storage VLAN configuration consistent across both fabrics. Check flow control or data-centre features only if the storage and switch vendors require them. Do not enable features from another vendor's design guide without confirming compatibility.

Packet loss

Storage protocols are sensitive to retransmission and latency. Watch interface errors, drops and oversubscription. If the array reports low storage latency while the application is slow, the host and network path deserve equal attention.

Maintenance test

Fail one path at a time during an approved window and watch application I/O, multipath state and recovery. A design that has never survived a controlled path failure is not proven redundant.

Discovery, login and presentation are different stages

An initiator can discover a target portal without successfully logging in, and it can log in without receiving the intended LUN. Troubleshoot the sequence separately: IP reachability, target discovery, authentication, session establishment, LUN mapping, multipath presentation and filesystem/application use.

Authentication

CHAP credentials, if used, belong to the iSCSI session. A successful ping does not prove iSCSI authentication will work. Keep initiator and target identity records so access can be audited.

Path ownership

Some arrays have preferred paths or controllers. Multipath software can use active/active, active/passive or vendor-specific behaviour. Follow the array's host integration guide rather than forcing a generic round-robin policy.

Storage VLAN routing

Many iSCSI designs keep storage networks non-routed to reduce complexity and exposure. If routing is intentionally used, account for MTU, latency, security policy and multipath design across every routed hop.

Fault example

If one path repeatedly drops while the other is stable, compare switch errors, optics/cabling, host NIC logs and target-port counters for that path only. Do not reset the entire storage system before isolating the failing leg.