No announcement yet.

Live Migration of Linux Guest Fails to Send a Gratuitous ARP and Loses Connectivity

  • Filter
  • Time
  • Show
Clear All
new posts

  • Live Migration of Linux Guest Fails to Send a Gratuitous ARP and Loses Connectivity


    I am having a problem with Linux guests that is hampering my ability to put a Hyper-V cluster into production. For Windows guests, everything works fine. Immediately after a live migration finishes, the guest sends a gratuitous ARP which takes care of setting up the FDB on the switch and updating the ARP table on the router. The latter is especially important as we are using Broadcom nic teams which rewrite the MAC of outbound packets with the MAC of the active nic team member. I have taken packet captures and can verify that the gratuitous ARP is sent. The guest also sends a couple of ARPs with a source protocol address of which I believe is a form of duplicate IP address detection.

    For a linux guest, no gratuitous ARP is sent. Therefore, the switch and router are not updated. Since the nic team is rewriting the MAC, all traffic is sent to the old host. If I ifdown and then ifup the interface, this forces Linux to send a gratuitous ARP and traffic is restored.

    I have verified that this is not a problem with the nic teaming software. I removed the teams on both hosts and repeated my testing. The gratuitous ARP is still not sent. Without the team, I don't have to worry about the ARP table on the router, but the FDB on the switch still has old data until traffic is sent outbound from the VM to refresh it.

    This is a serious problem for me, as I need to be able to host functional Linux guests. I have not seen mention of this problem anywhere. I have seen lots of people complaining about the Broadcom teaming software, but I have verified this is not the cause of the gratuitous ARP problem. Furthermore, everything works great with windows guests.

    My working theory is that Hyper-V is instructing the virtual ethernet device to send a Gratuitous ARP after a live migration completes, and that there is a bug in the Linux integration components that prevents this from happening. I am currently testing with Centos 5.5 running version 2.1 of the Linux IC. My hosts are Server 2008R2 with all of the latest patches and recommended hotfixes. I have tested with both emulated and synthetic adapters. The adapter is connected to an external network, and the management operating system is not allowed to share that network. A VLAN is tagged by Hyper-V.

    I hope there is someone who can assist with this, or even verify that they have been able to repeat the problem.

  • #2
    Re: Live Migration of Linux Guest Fails to Send a Gratuitous ARP and Loses Connectivi

    I've seen one other person on a MS forum with the same problem. Has no one else experienced this? Perhaps many are having this issue and don't realize it as a VM with sufficient outbound traffic will reset the FDB on the switch as soon as a packet hits it.

    Since I am running Broadcom teaming software which rewrites the MACs, I have to worry about the ARP table on the router stuck with the wrong mapping.

    Can anyone using nic teaming with Intel nics chime in on whether it is possible to configure an Intel nic team which does not rewrite the MAC on virtual machine packets in the way that the Broadcom teaming software does?