VM vMotion is failing despite confirmed ICMP connectivity between the vMotion VMkernel interfaces
Article ID: 427223
Updated On:
Products
Issue/Introduction
Attempts to perform a vMotion migration between ESXi hosts fail consistently.
While initial troubleshooting via Testing VMkernel network connectivity with the vmkping command shows successful connectivity between the source and destination VMkernel ports, the migration task times out or fails during the initial handshake phase .
vMotion requires a stable, dedicated TCP connection between the source and destination ESXi hosts. Even if basic ICMP (ping) traffic appears functional, underlying network anomalies—specifically those involving ARP (Address Resolution Protocol)—can prevent the establishment of the reliable TCP stream necessary for data transfer.
In this scenario, while the IP address responds to pings, the underlying MAC address mapping is inconsistent, leading to connection loss.
Environment
VMware vSphere ESXi
Cause
In such cases the failure is caused by an IP Address Conflict on the vMotion network.
An end to end packet capture, taken on the vnic, vswitch and uplink layer can be used to prove this IP address conflict.
Consider the vmotion of VMs happening from a source host (x.x.x.96) and destination host (x.x.x.112).
Captures on source Host (x.x.x.96) :
Captures on destination Host (x.x.x.112) :
Packet 2 (x.x.x.96) : A broadcast packet is sent as a broadcast packet to know the destination IPs (Who has x.x.x.112? Tell x.x.x.96)
Packet1 (x.x.x.112) : The broadcast packet is received.
Packet5 (x.x.x.112) : The host replies with the MAC associated with its IP address (27:c8)
Packet3,4,5 (x.x.x.96) : In the reply same host IP is seen, but with a different MAC (65:28)
This shows that there exist two different MAC addresses (27:c8) and (65:28) mapped to the same IP address i.e x.x.x.112
Resolution
A unique IP address should be assigned to this vmkernel for vmotion.
Feedback
