Fibre Channel HBA I/O load imbalance on ESXi 8.x

Products

• VMware vCenter Server

Issue/Introduction

Users may observe a significant I/O load imbalance between redundant Fibre Channel Host Bus Adapters (HBAs) on ESXi 8.x hosts. Symptoms include:

• One HBA processes significantly more I/O (e.g., a ratio of 9:1 or higher) than its peer.
• All storage paths report as "Active" or "Healthy" in the vSphere Client.
• esxtop shows zero or minimal throughput on one HBA while the other is heavily utilized.
• Performance degradation or latency on virtual machines.

Environment

 

• VMware ESXi 8.x
• Fibre Channel Storage (e.g., Pure Storage, Dell PowerStore, NetApp)
• VMware Native Multipathing (NMP)

 

Cause

This behavior occurs when the ESXi storage stack detects underlying physical layer instability. Specifically, if an HBA path experiences "Link Flaky" events, the ESXi Native Multipathing (NMP) driver dynamically avoids routing I/O to that path to prevent latency spikes and protect data integrity.

Common triggers for "Link Flaky" states include:

• Faulty SFPs on the HBA or SAN switch port.
• Degraded or contaminated fiber optic cables.
• SAN fabric congestion or Buffer-to-Buffer (B2B) credit exhaustion.

Resolution

To resolve the imbalance, identify and mitigate the physical layer instability:

1. Review the /var/run/log/vmkernel.log for "Link Flaky" or "Performance has deteriorated" messages associated with the underutilized HBA.
2. Validate HBA statistics using the following command: esxcli storage san fc stats get -a vmhbaX Look for increments in Loss of Signal Count or Invalid Tx Word Count.
3. Engage your SAN/Fabric vendor (e.g., Brocade, Cisco) to inspect switch port telemetry for frame drops or CRC errors.
4. Replace the SFP module and fiber optic cable for the affected HBA port.
5. Once the physical link is stabilized, VMware NMP will automatically resume balanced I/O distribution across all healthy paths.

Additional Information

For further diagnostic steps, refer to .

Subscribe to this article to receive updates on fix status: .