iSCSI Cabling and LACP Mistake on a New PowerStore 500T
A new PowerStore 500T is being installed at a branch office with two Cisco Catalyst 9500 switches, SW1 and SW2, dedicated to iSCSI. Each node has two 25GbE ports for host I/O. The installer cabled node A port 1 and port 2 to SW1, cabled node B port 1 and port 2 to SW2, and configured a two-port LACP bond on each node to double bandwidth. The Windows Server 2022 hosts each have two 25GbE NICs, one connected to SW1 and one to SW2, and use MPIO. During validation, each host sees only one path to each volume and aggregate throughput tops out near 2 Gbps, far below the expected 20 Gbps. The switches are interconnected with a 100GbE ISL. The customer wants to fix this without re-cabling every host.
1. What is wrong with the array cabling and LACP design? Explain the effect on multipath.
Read the worked answer
Two separate mistakes are present. First, PowerStore iSCSI host ports are independent interfaces used for multipath I/O; they are not meant to be combined into an LACP bond for host access. Bonding them changes the expected many-to-many path design and can make the host see fewer target portals than intended. Second, both ports of node A are on SW1 and both ports of node B are on SW2, so a single switch failure removes an entire node from the host's path set. That defeats the purpose of dual switches.
The host MPIO layer expects multiple independent target IP addresses, ideally reaching both nodes through both switches. With the current design, the host may see only one effective portal or one node's paths, which explains the single path and the low throughput. The 100GbE ISL does not fix the loss of redundancy at the node-to-switch level.
2. Recommend a corrected cabling and IP design using the existing two switches, and describe host-side changes.
Read the worked answer
Re-cable so that each node has one 25GbE host port on SW1 and one on SW2. For example, node A port 1 to SW1 and port 2 to SW2; node B port 1 to SW1 and port 2 to SW2. Do not configure LACP on these ports. Give each array port its own IP address on the storage VLAN, and make sure the two switches are Layer 2 adjacent for that VLAN through the ISL. If the design uses separate storage subnets per switch, keep each port on the correct subnet and verify routing and MTU.
On the Windows hosts, keep one NIC on SW1 and one on SW2, enable MPIO with a supported policy such as round robin or least queue depth, and rescan. The goal is for each host to see four target portals across two nodes and two switches, with no single switch or node failure removing all paths. Do not use teaming/LACP for the storage NICs; use MPIO.
3. What would you check in PowerStore Manager to confirm the fix?
Read the worked answer
Check the hardware view for each node's 25GbE port link status and speed, and confirm that each port has the intended IP address and is not bonded. In the iSCSI interfaces view, verify that all four host-facing interfaces are up and reachable. On the host side, confirm that the initiators are logged in and that each host sees the expected number of target portals. In PowerStore Manager, review the host's initiator records and the volume mapping to ensure paths are present from both nodes.
Finally, run a quick I/O validation from the host and check PowerStore performance metrics for iSCSI bandwidth and latency across both nodes. If only one node shows traffic, the problem is still cabling, IP, or host MPIO configuration, not array performance.